Science topics: Recognition
Science topic

Recognition - Science topic

Explore the latest questions and answers in Recognition, and find Recognition experts.
Questions related to Recognition
  • asked a question related to Recognition
Question
2 answers
I want to work on emotion recognition data that will be analyzed on few subjects for an offline analysis. I have heard about OpenBCI a the most popular option but I do not think that it is cheap at $3400+ . I wanted to check if Muse and Neurosky that have a sampling rate of 256Hz/12bits are decent enough to report a study acceptable in good SCI/SSCI journals ?
  • asked a question related to Recognition
Question
3 answers
As most of us know, the reviewer recognition website PUBLONS was taken over by CLARIVATE (Web of Science) last year and, a few months ago, Publons was merged with WoS., thus factually disappearing.
What do you think about this new scenario? Are we reviewers sufficiently recognized in WoS website? I personally loved the fresh, clear and dedicated webpages from Publons, and I don't think WoS communicates reviewer recognition as straight and clear as Publons did.
I would like to hear your opinions about this matter and about how to improve reviewer recognition in general. What can be done to improve the public recognition of our contribution to the publishing process? Every publisher says our hidden work is "crucial" to science communication, however I am not sure we are rewarded as we deserve...
Relevant answer
Answer
Yes, Linas Balciauskas and Thomas S. Jung you both are OK. Publons was great. But what can be done to better recognise the work of reviewers? PUBLONS is as good as gone, so what comes next? ORCID has a section devoted to one's peer review activity. But is that enough? Not for me, anyhow...
  • asked a question related to Recognition
Question
1 answer
Hello everyone. I am working on a paper on facilitating affordance perception in e-learning design. I read a paper on emotional affordances and how they impact students' emotions in online learning. My question is does emotions or emotional affordances play a part in the recognition process of functional affordances since perception is a recognition process according to TACT theory?
Relevant answer
Answer
Hello Clement,
Students emotions in online education is an essential topic as affective procedures influence cognitive procedures of learning [1]. Are you focusing on automatic recognition or students emotions analysing text [2] or on the emotional enagement of students using e.g. motivational enhancement methods such as stories, narration, playful or gameful design [3]?
[1] Mystakidis, S. (2021). Deep Meaningful Learning. Encyclopedia, 1(3), 988–997. https://doi.org/10.3390/encyclopedia1030075
[2] Tsimaras, D. O., Mystakidis, S., Christopoulos, A., Zoulias, E., & Hatzilygeroudis, I. (2022). E-Learning Courses Evaluation on the Basis of Trainees’ Feedback on Open Questions Text Analysis. Education Sciences, 12(9), 633. https://doi.org/10.3390/educsci12090633
[3] Mystakidis, S., Filippousis, G., Tolis, D., & Tseregkouni, E. (2021). Playful Metaphors for Narrative-Driven E-Learning. Applied Sciences, 11(24), 11682. https://doi.org/10.3390/app112411682
  • asked a question related to Recognition
Question
1 answer
Is it possible to use Cas9 (WT) to perform CRISPR-screening with a sgRNA subpooled library (Addgene catalog number: 83971) designed to be used with dCas9 systems? Beyond mutations at nuclease domains (HNH and RuvC), is there any other difference between Cas9 and dCas9 enzymes that could lead to impairment of sgRNA recognition by Cas9 (WT)?
Thanks!
Relevant answer
Answer
There isn't a difference in the sgRNA binding between Cas9 and dcas9.
One thing you need to check is what is the library targeting, libraries designed for dCas9 are likely intended for CRISPRa/i type screen and as such are likely to target promoter which may not be optimal for a mutation based screen.
  • asked a question related to Recognition
Question
4 answers
I'm working on a project about palm vein recognition, and I need a dataset of palm vein images without a license agreement.
Relevant answer
Answer
Among the members of biometric identifiers, the palmprint and the palmvein have received significant attention due to their stability, uniqueness, and non-intrusiveness. In this paper, we investigate the problem of palmprint/palmvein recognition and propose a Deep Convolutional Neural Network (DCNN) based scheme, namely P a l m R CNN (short for palmprint/palmvein recognition using CNNs). The effectiveness and efficiency of P a l m R CNN have been verified through extensive experiments conducted on benchmark datasets. In addition, though substantial effort has been devoted to palmvein recognition, it is still quite difficult for the researchers to know the potential discriminating capability of the contactless palmvein.
Regards,
Shafagat
  • asked a question related to Recognition
Question
5 answers
We have aerial survey photos taken by a drone. Images are already in GPS - raster format. I would like to analyze (locate and count) structures - specific bird nests in reedbeds. Which would be the best software to do this, either self standing or ArcGIS/QGIS add-on. It would be some kind of software for shape recognition by machine learning.
Relevant answer
Answer
Check the module "Feature Extraction" by ENVI
  • asked a question related to Recognition
Question
4 answers
Basically one needs to get a journal article published to get recognition of one's work. How does one go about achieving this promptly and successfully?
Relevant answer
Answer
The answer to your question is very hard but important and to me, it depends on the requirements of your employer or institution. It depends on their requirements. Do they want it to be a published in a certain type of journal with a certain indexing or classification eg Pubmed/Medline. That would make it a bit harder. If that is not required specifically, it gives you a lot more options in some fields and most commonly, there are faster publishing times with some journals when you pay for submission or publication.
  • asked a question related to Recognition
Question
4 answers
Databases doesn't accept a duplicate entry but what about a similar entry of two similar photos can the deep face recognition figure this out and how about manipulation in the photo like flipping changing brightness and contrast can this affect the recognition ?
Relevant answer
Answer
Applying machine learning techniques to biometric security solutions is one of the emerging AI trends. Today I would like to share some ideas about how to develop a face recognition-based biometric identification system using OpenCV library, DLib and real-time streaming via video camera.
Regards,
Shafagat
  • asked a question related to Recognition
Question
6 answers
Hello,
Can anyone share with me a Health-dataset that is related to emotion recognition, datasets that contents of Text, Audio, Visual, and physiological?
Relevant answer
Thank you so much,
  • asked a question related to Recognition
Question
6 answers
I want to compare recognition rate with other dataset, so that  I need  Isolated printed Arabic character datasets.
  • asked a question related to Recognition
Question
3 answers
During the recognition, It has a process of information matching. How does it match successfully despite some variation?
Relevant answer
Answer
Unlike the machine (computing devices) which goes pixel by pixel, the brain performs gestalt recognition. Based on (top-down) feedback from the brain, the eyes quickly focus on only what is the main object of interest and increases the resolution for the same. Based on its past learning of possible variations to other objects (in terms of views, noise, etc.), it makes an informed guess of the object even though there may be variations unseen before. If the accuracy of recognition is confirmed, the brain immediately learns the new pattern also, with its variation. However, with computer vision, you need to explicitly invoke the training session to add new patterns.
  • asked a question related to Recognition
Question
7 answers
Please let me know the name or URL of any comprehensive Bangla corpus data for SA or ER.
  • asked a question related to Recognition
Question
4 answers
I have a datasets which is comprised of skeleton XML files and RGB videos, I want to apply classification, my question is how we can use both data features as whole input?
I have seen papers based on RGB+Depth data.
Please direct me.
Thanks
  • asked a question related to Recognition
Question
6 answers
I'm new in CNNs. After discovering the pre-formed deep CNNs, which can be used for feature extraction, I ask myself if there is still an application area in the field of image recognition, in which the classical classification methods (such as e.g. SVM etc.) are preferred to the CNN's?
Relevant answer
Answer
A big yes, SVMs or any other classical non deep architecture are still used, you should use them when you are confident that the extracted features are of good significance. You can also use a CNN as a feature generator for a downstream SVM based classifier.
  • asked a question related to Recognition
Question
4 answers
Hi,
I have been working on InHard Dataset, human action recognition for human robot collaboration. I use different machine learning algorithms on the dataset for segmented action recognition but the results I got is very poor. How can I publish results? There is no work on validity of dataset, just a single paper is published on the dataset description. Can anyone please guide me? Or please share any paper that did similar work (Poor results).
Relevant answer
Answer
A research outcome may have different dimensions. It can be either positive, negative, or both.
What matters most is the contribution of the research.
If the outcomes of your research make a significant contribution to the relevant domain, then whatever the result is, it can be publishable.
  • asked a question related to Recognition
Question
5 answers
Hi every one,
I'm a PhD student and I'm currently working on iris recognition.
I already investigated a new method of iris segmentation that you can find in this link : (http://www.sciencedirect.com/science/article/pii/S0957417416306741).
I need to know is there any method to create an electronic system or electronic card where i can implement my segmentation method ? card type, programming language, ...etc.
Thanks you.
Regards.
Relevant answer
Do you applied segmentation with threshold tecniques using the threshold and later mathematical morphology ?
  • asked a question related to Recognition
Question
1 answer
I did an experiment on emotion recognition in Psychopy. 60 videos were shown. At the beginning and end of each video, a trigger signal was sent from Psychopy to Biopac. This trigger signal indicates when the video starts and when it ends. Now to my question: In the analyze in Biopac I want to fix only the time interval in which the trigger signal was present (i.e. only the time in which videos were shown). How can I automatically fix these intervals in which the trigger signal is there?
How can I fix fixed intervals that are then analyzed? I want to fix exactly the intervals in which the trigger signal is present. Does anyone know more about this ?
Relevant answer
Answer
First, convert digital inputs to stimulus events. ("Analysis" - "Stim Response" - "Digital Inputs to Stim Evetns")
This marks "light bulb icons" at the beginning and ending points of your videos.
Then, find cycles using the light bulbs you've created. ("Analysis" - "Find Cycle" or Ctl+F.
This finds events you've marked on the graph.
("Cycle/Peak" - "Events" - "Start event: Stimulus Delivery; located on Global" - End event: Stimulus Delivery: located on Global")
Finally, save the outputs pulled from the highlighted cycles. Before doing this, physiological measures such as NS-SCR should be analyzed BEFORE finding cycles on the graph.
The only problem that I can see is that the beginning and ending points of the videos are marked on the same channel. Since Acqknowledge indistinguishable identifies the beginning and ending points as the same "stim events (light bulb icons)," not only "beginning to ending" intervals but also "ending to another beginning (I believe this is NOT what you want)" intervals will be highlighted as the cycles of your interest on the graph.
However, if the intervals between videos (ending to beginning) are longer/shorter than the duration of the videos (beginning to ending), you'll be able to filter those out AFTER getting the full outputs.
In summary, although the digital inputs of your data have been set up in a less ideal way, the above solution might still work for your data.
If you're more interested in "event-related" physiological responses (e.g., understanding how people responded differently to the different types of videos), the way you've set up your digital inputs is okay as it is now. The only thing you need to be careful of is to define the minimum and maximum separation rates appropriately (maximum separation rate is typically 8-10 seconds).
More information about the difference between non-specific and event-related analyses can be found in the following manual.
I hope this helps!
  • asked a question related to Recognition
Question
4 answers
Hello!
I have a sample health dataset that contains clinical notes and descriptions. I have applied for Name Entity Recognition (NER) over the dataset but do not feel the need for any information such as age, gender or address. I would like to apply the de-identification technique on top of the patient's clinical data. Would you please share with me some resources where I can get information on the implementation of de-identification using Python?
Thank you
Relevant answer
Answer
Thank you very much for this information, and I appreciate your valuable time.
  • asked a question related to Recognition
Question
4 answers
Hi,
does anybody have information that can share about pain recognition in bats?
Citable sources will be especially appreciated.
Thank you
Javier
Relevant answer
Answer
Hi Carl
thank you for your suggestion. I am aware of these books, but I don't have access to them, unfortunately.
Javier
  • asked a question related to Recognition
Question
3 answers
Does anyone get errors with using the Recognition part in " Scene Text Detection and Recognition by CRAFT and a Four-Stage Network" ?
Relevant answer
Answer
etecting and Recognizing texts for a given natural scene image using EAST and Tesseract Algorithms.
Regards,
Shafagat
  • asked a question related to Recognition
Question
4 answers
If any technique works over odia then please suggest it here.
Thanks in advance.
Relevant answer
Answer
please look at the following link:
  • asked a question related to Recognition
Question
5 answers
Hi,
I want to train a model for real time robot, I have a set of videos, want to train model for action recognition and model will be fed to robot to identify human actions. I never worked before for on video data. How should I start?
Relevant answer
Answer
Just so that I understand your setup correctly could you let me know the operating system being used? The reason is that all peripheral devices are controlled by the operating system kernel.
For example in Linux (which is what I use) I can communicate directly with the OS video interface (V4L) and do a program in C that extracts the frame in YUV format
An example of the C program is:
note that in the function init_device the resolution is set along with the format
fmt.fmt.pix.width = 640; fmt.fmt.pix.height = 480; fmt.fmt.pix.pixelformat = V4L2_PIX_FMT_YUYV;
With this properties and the read_frame function allows you to extract the particular frame and the process_image sends it to standard output. Instead of sending it to standard output you would need to send it to a function that converts the YUV signal to the required format for processing by your routines. The function will have to convert each of the values of YUV for the resolution specified to RGB in values from 0-255. the conversion in your function would need to do a translation to RGB using the formulas in:
Once you do this you will have a buffer that is 640*480*3 (the values above times the the R, the G, and the V values). If you just want greyscale array that is 640*480 you can use:
Grayscale = (R + G + B ) / 3.
then you will just have a buffer that is 640*480
At this point you can then use the frame for processing by the ML or you can resize it to the appropriate resolution.
This is just to give you the general idea. The same steps will have to be done in Python and whether it is Windows or Linux.
Note that some libraries will manage this process and convert them to AVI or MP4. Due to your real time constraints I would suggest that you do not waste processing time converting them and then decode them and waste time parsing the format structure.
Let me know
  • asked a question related to Recognition
Question
7 answers
A question about models of recall and recognition based upon personal experience:
I have had the experience of not being able to recall a specific words (I hate growing older ;) ), but still can pull up enough context to do a search for the word. But here has been my experience--before I begin to scan the results of the search I suddenly am able to recall the word; it is almost as if the mere act of beginning the scan there is some level of 'pre-activation' that enables me to recall, but it feels as a gray area between recall and recognition. Now, I have not been active in cognitive psychology for a while, but my quick research review has not provided insight. So for those cognitive psychologists out there, including developmental psychologists, as it seems to be there are developmental processes that must be at play here, is there a body of specific to these processes? Perhaps it is just me and not a more global process at all?
Relevant answer
Answer
I think the phenomenon you observed in yourself is not only common but has already been observed by Aristoteles, however it has been summarized in more specific theoretical frameworks in the 20th century.
Best,
Sven
  • asked a question related to Recognition
Question
1 answer
Dear colleagues,
I am struggling to get an acceptable Cronbach's alpha score for a scale on opportunity recognition. Opportunity recognition was measured using Nicolaou, Shane, Cherkas, & Spector’s 5-item scale (2009) with the answer categories ranging from “strongly disagree” to “strongly agree.” These scale questions are drawn from the literature on opportunity recognition (Ozgen & Baron, 2007; Singh et al., 1999). In order to get a Cronbach's alpha score that is acceptable, I need to eliminate items 4 & 5. How do I justify that in my write-up? Is there a better way to handle that?
Relevant answer
Answer
One way is to remove the outliers and recalculate the Cronbach. I am sure you have done that. The second method is to modify those items, do a pre-test followed by an FGD with the same respondents, and then pilot it for the reliability analysis.
  • asked a question related to Recognition
Question
7 answers
In the Domain of visual recognition tasks if imbalanced data is there our model gives poor performance?
Relevant answer
Answer
Interesting suggestion thanks a lot all!
  • asked a question related to Recognition
Question
3 answers
Transformation of industry economy
Relevant answer
Answer
Three factors are vital & these are organization factors, technology factors and management factors.
  • asked a question related to Recognition
Question
6 answers
What is optimal feature selection algorithms in visual recognition tasks?
Relevant answer
Answer
You can check these articles; they will help you to understand this.
  • asked a question related to Recognition
Question
2 answers
Following up on the previous question, linguists are still disputing how languages are organized in bilinguals’ minds: whether a speaker’s native and second languages are separate entities or whether they form an entity in their minds. Nobody can deny that both schools had empirical evidence to back up their claims. On the other hand, many linguists argue that the location of the language has little bearing on the production and recognition of words. The relationships between words in the mental lexicon are more important. Do you agree with this viewpoint?
Relevant answer
Answer
Given that the issue is the structure and organization of languages in the minds of bilinguals, the location of language in this context refers to whether L1 and L2 are located together or independently in a certain area of the mind. As mentioned in the discussion, there is a considerable body of literature on the topic, but no agreement has been reached. In addition, some scholars claim that this issue is peripheral, as it isn't important to the production and recognition of lexical units.
  • asked a question related to Recognition
Question
3 answers
I have a project in which I have given a dataset (more than enough) of 10-20seconds audio files (singing these "swar" / "ragas": "sa re ga ma pa" ) without any labels, nothing well in data ... and I have to create a deep learning model which will recognise what speech it is and for how long it is present in the audio clip (time range of particular "Swar" sa ,re ,ga, ma )
The answesr to questions that I am looking for are
1. how I can achieve my goal , should I use RNN , CNN ,LSTM or hidden Markov model or something else like unsupervised learning for speech recognition ?
2. How to get correct speech tone for Indian language as most acoustic speech recognition models are tuned for English ?
3. How to find the time range ?for what range particular sound with particular "swar" is present in music clip ? how to add that time range recognition with speech recognition model ?
4. are there any existing music recognition models which resembles my research topic ? ,if yes please tag them .
I am looking for full guide for this project as it's completely new and people who are interested to work with me /guide me are also welcome .
  • asked a question related to Recognition
Question
3 answers
I am working on phoneme recognition. Is there any dataset of sound files for each phoneme class?
  • asked a question related to Recognition
Question
10 answers
We are really curious to see if there is anyone else here who is working on Artificial Intelligence and anxiety? If you are, please share your work here. Thank you! Here are some of the work we recently completed and some of the media coverage our work got:
#Artificial intelligence
#Anxiety
#Anxiety Measurement
Relevant answer
Answer
Researchers are using artificial intelligence (AI) to detect behavioral signs of anxiety with more than 90 percent accuracy, and suggest that AI could have future applications for addressing mental health and well-being. Their research is published in the journal Pervasive and Mobile Computing Gulnaz Anjum
  • asked a question related to Recognition
Question
2 answers
In speech based emotion recognition, for individual artists almost get 90 percentage of accuracy using neural network, but accuracy is decreased up to 70 percentage for entire speech corpus.
How can I increase accuracy?
My speech database contain 10 artists.
Relevant answer
Answer
Interesting topic.
  • asked a question related to Recognition
Question
7 answers
Is the South African Department of Basic Education providing enough assistance and recognition to student teachers who are preparing to become future teachers in South Africa?
  • asked a question related to Recognition
Question
3 answers
Please any one tell me
Relevant answer
Answer
I think Convolutional Neural Network CNN and its variants are dominant in this topic.
  • asked a question related to Recognition
Question
4 answers
For many smart road applications, objects detection and recognition are one of the most important components. Indeed, precise detection of road objects is a critical task for autonomous urban driving and robotics technologies.
Actually, we are working on the conception of a smart system that consists in detecting and blurring undesirable road objects to anonymize and secure road users.
In this context, we want to know what are the up-to-date methods (Neural approaches, etc.) used for the detection and recognition of road objects (humans, vehicles, license plates, etc.) ? Any ideas ?
Greatly appreciate your review
Relevant answer
Answer
Dear Prof. Wissam Kaddah,
I have the following research in this area:
  • asked a question related to Recognition
Question
8 answers
Hi, I am new (next to zero experience) to circular dichroism. Can anyone tell me what secondary structure is associated with the following CD spectra? After some reading, I found this article and assume it is anti-right-twisted beta strand. If anyone agree or otherwise, please explain, and maybe you could also tell me how to properly read a CD spectra.
Thanks!
Relevant answer
Answer
Looking at the resemblance it might be, but things like aggregation might explain the low signal at shorter wavelengths and could lead to another conclusion. Personally, I would check whether the CD analysis makes any sense, the sequence of your protein or peptide can be used for structure prediction, there are numerous programs for this. Obtain the so-called FASTA form and start with prediction methods:
-First try secondary prediction methods like PsiPred. It is common practise to use at least two different methods:
SOPMA:
Porter5:
JPred4:
-Also try (provided that there is no X-ray structure) 3D prediction might be considered, I-TASSERis one of the best methods for this purpose (and result in a pdb file which can be used for further analysis). See for an example:
Identification of Possible Lipid Binding Regions in Food Proteins and Peptides and Additional In Silico Analysis
-Possible alpha-helix regions can be further analysed by Heliquest and renders helical wheel plots and all sorts of physicochemical data (like an idea on whether you deal with a hydrophobic peptide/protein region):
Article HELIQUEST: A web server to screen sequences with specific α-...
One need to keep in mind that predictions have a certain degree of uncertainty (or put in more positive words an accuracy of round 90%), so intrinsically it is possible that your protein is an exception, and the predictions are not entirely accurate. Therefor experimental confirmation is necessary for example like your circular dichroism approach. However, ‘simply’ put protein in buffer and measure might not always render relevant data. Possible influence of difference in variables like pH, ionic strength, temperature etc. need to be checked. Also, for example the absence or presence of anionic phospholipids can dramatically affect the secondary structure, see for example:
Article Anionic phospholipids are essential for α-helix formation of...
Chapter Characterization of helical structures in gelatin networks a...
In the ideal world the predictions and experimental data are in line. You might have to consider other tools like FTIR, NMR, ESR etc. as well.
Bet regards.
  • asked a question related to Recognition
Question
3 answers
I would like to receive suggestions about which indicators can better represent the recognition that non-state armed groups can receive from the international community.
Relevant answer
Answer
عقد اتفاق مؤشر للاعتراف
  • asked a question related to Recognition
Question
5 answers
We can connect you with fellow researchers. We provide fair pay and recognition for supporting your peers.
Become a Peerwith expert
Relevant answer
Answer
I appreciate this great effort, and I hope to join you, but it is possible to know a biography of the work, as well as your presentation of a comprehensive plan on the most important achievements and humanitarian psychological works that serve patients or those who need them, because as you know, we have a duty towards the right of every patient. Therefore, we are obligated to offer our sincerity to work before God Almighty, and before those who need guidance and psychological treatment.
  • asked a question related to Recognition
Question
3 answers
Hello everyone,
I am just working on my master thesis and for this I am planning to find out the localization of a gene, transfected into HEK293 cells. One idea of mine was to digest the genomic DNA with the enzymes SspI, NheI, SpeI and AvrII. The known sequence of my insert will be digested by SspI and on the other hand digest the unkown genomic sequence with NheI, SpeI and AvrII, which produce compatible cohesive ends. In the next step I want to ligate linkers to the sticky ends, and perform a PCR on this template with the primers binding in the linker, so that finally the genomic DNA gets amplified.
To see whether this setup could work I wanted to perform a model experiment on an already modified HEK293 strain, which already obtained the SspI recognition site and the localization of the insertion site is known. But when I checked the genomic flanking sequences, there were no recognition sites for NheI, SpeI and AvrII within the downstream 5000 bp and a subsequent PCR is not really possible. Accordingly, I think my setup for the genome walking is not optimal chosen. Has anybody an idea which restriction enzymes can lead to appropiate results?
Thanks for the help.
Relevant answer
  • asked a question related to Recognition
Question
7 answers
For example AI can automate processes in the initial interpretation of images and shift the clinical workflow of radiographic detection, management decisions. But what are the clinical challenges for its application?
#AI #cancer #clinical #oncology #biomedical
Relevant answer
Answer
The first is the sample size issue, a predictive model are trained and built with limited imaging data from a particular hospital or region; this has to face bias caused by various factors when dealing with a wider range of diagnoses; it may not be adaptable when making diagnostic decisions for images from other regions.
The second is the inability to balance the patient's treatment intentions, the patient's financial situation, and different regional customs when making treatment decisions.
  • asked a question related to Recognition
Question
5 answers
  1. HAR dataset - human activity recognition I have used.
  2. What is the impact of activity recognition if the dataset is having small number of features?
Relevant answer
Answer
You may encounter over-fitting which may impede accurate identification of unseen features.
  • asked a question related to Recognition
Question
5 answers
If you have a recommendation of someone and a summary of the importance of their contribution to science, I would like to learn of these great researchers.
I ask because I found out only recently that my very first research advisor, Dr. Bob Behringer, had passed. He was an amazing, personable figure in granular materials physics and his story is here: https://today.duke.edu/2018/07/physics-professor-robert-p-behringer-dies-69. I was recommending his research to a colleague who had a question about force chains and that's when we learned the news. He was only 69 but he had been a part of the Duke University community for nearly 50 years.
It was too late for my colleague to ask Dr. Behringer his question, but maybe this discussion will help raise awareness of some influential research from some great people. Maybe we can reach out to them with our questions before it is too late. And if it is too late because they have recently departed, their work will live on through those who know about their research.
Relevant answer
Answer
your idea is perfect. the world is full of these wonderful people that are expert on their works but I feel sorry that I can't become one of the student of them.
for everyone that works on seed topics, I'd like to introduce prof Gerhard Leubner in Royal Holloway, University of London who I have cheched his works for a long time and I love him by my heart .
  • asked a question related to Recognition
Question
11 answers
Dear colleagues, I am working on a project of Object detection with Bounding box regression (on Keras, TensorFlow), but I can't find decent sources and code samples. My task involves 200 classes, and each picture can contain from 10 to 20 classes that need to be detected. A medical task, more precisely, the recognition of types of medical blood tests.
I fixed some works that are close to mine. But I want to implement my task for the Russian language and on the KERAS & TensorFlow framework. I am sincerely grateful for any advice and support!
Relevant answer
Answer
  • asked a question related to Recognition
Question
8 answers
I searched for information about pattern recognition, which means creating a pattern of output data from devices such as each sensor or each actuator or single machine in factories. but most of the studies are about users recognition behaviour patterns. I want to have a behaviour pattern from each device in the factory.
can you advise me, please?
Relevant answer
Answer
Look the link, maybe useful.
Deleted research item The research item mentioned here has been deleted
Regards,
Shafagat
  • asked a question related to Recognition
Question
2 answers
  1. I have worked on HAR and HAPT UCI datasets for human activity recognition which has 561 features.
  2. But with a huge dataset of low dimensional space (2 or 3 dimension), can we do activity recognition and what will be its impact?
Relevant answer
  • asked a question related to Recognition
Question
22 answers
Last Tuesday (8), exactly on World Oceans Day, the US National Geographic Society cartographers decided to present planet Earth with another ocean by adding the Southern Ocean, which surrounds Antarctica, to its four "brothers" older Atlantic, Pacific, Indian and Arctic.
According to Seth Sykora-Bodie, a marine scientist at the US National Oceanic and Atmospheric Administration (NOAA), while it's hard to explain what's different about the region, “everyone will agree that the glaciers are bluer, the air cooler, the mountains more intimidating and the landscapes more captivating than anywhere else you can go,” he told National Geographic.om another ocean by adding the Southern Ocean, which encircles Antarctica, to its four older “brothers” Atlantic, Pacific , Indian and Arctic.
NOAA's official geographer Alex Tait also spoke to the NatGeo website, explaining that official recognition just didn't happen before because “there was never an international agreement”. “It's kind of a geographic geek in some ways,” he summed up. But as of June 8, 2021, the nerd is over: the Southern Ocean is officially the fifth ocean on Earth.
Although the difference between sea and ocean is clear, the size, it is still difficult to say exactly what an ocean is, other than to repeat that they are more extensive liquid masses (the smallest of them is 73 million square kilometers) and deeper. The Spanish navigator Vasco Nuñez de Balboa already recognized, in the beginning of the 16th century, that those waters at the bottom of the world were the "Antarctic Ocean".
Relevant answer
Answer
Good question! but it is out of my field.
we follow to learn from experts. Good luck!
  • asked a question related to Recognition
Question
3 answers
I would like to used to use sumo protease. However, I did not find the sequence that the protease recognizes in order to order a plasmid with this sequence? and the information available on the internet using these sequences for many other purposes that make me confused about the function of this protease as a protein fusion domain ??
Anybody has ever used the SUMO as a cleavage protease? it is someone who did could you please share with me the recognition sequence or the paper, please. Thanks
Relevant answer
Answer
The SUMO-tag/SUMO protease system is special because it does not have a particular short recognition sequence (like other proteases) for the protease but the whole SUMO protein is needed to make the protease cut at the Gly-Gly-site at the end of SUMO right before the first amino acid of the target protein (can be any amino acid except for proline). You can contact the EMBL Protein Expression and Purification Core Facility where you can get an expression vector.
  • asked a question related to Recognition
Question
5 answers
Relevant answer
Answer
Check some journals in MDPI, Taylor and Francis, Springer, etc.
  • asked a question related to Recognition
Question
5 answers
Hello everyone,
I am working on a project which includes object recognition, currency recognition, text to speech, and location of the user. In order to perform the currency recognition of Pakistan's currency, I'd need a lot of data and computing resources to train a model, but unfortunately, I don't have access to either of these.
So, I just wanted to know whether there is an open-source pre-trained model that I could use for my project?
Any help would be appreciated.
Thank you.
Relevant answer
Answer
  • asked a question related to Recognition
Question
8 answers
What is the best way to take the features of image i.e. edges, shapes, corners included in Character recognition?
Relevant answer
Answer
Hi
there is no direct algorithm used for all applications
But as an example you can apply DCT and DWT to extract features.
Regards
  • asked a question related to Recognition
Question
10 answers
In the early childhood music education, The reading ability of piano score will affect the learning ability and interest. Are there any effective methods or activity that can be applied to general teaching activity?
Relevant answer
Answer
How about designing an APP, which is basically a game, connected to a MIDI keyboard, wherein, if the player hits the right keys, the game progresses, while playing the correct music.
Similar to this, only for 2 hands, and with the right melodies (could input MIDI)
Also more examples here:
  • asked a question related to Recognition
Question
2 answers
One interpretation includes the following explanation:
Application of computer science and technology, special purpose scanners, for recognition of signals obtained by excitation of magnetic fields in electrotechnical devices, apparatus and measurement.
Relevant answer
Answer
Vishnu Kumar Gupta's
Additional question:
What does UDC 004.6:004.86 mean?
  • asked a question related to Recognition
Question
6 answers
What is the potential of blockchain technologies (DLT) in the field of e-government, and in particular for procedures for recognition of credentials? If so, what are the specific hurdles in implementing a DLT-based solution in the context of e-administration at transnational level? I recently published a research article on blockchain, e-government and recognition which you can access and download at: (published by Elsevier, in Research in Globalization).
The fact is that beyond bitcoins and some Fintech applications, DLT / BC has led to encouraging use cases in some transnational administrative procedures, e.g. recognition of qualified exporters in the context of customs procedures, or in the field of government procurement bids. But for now that's about it. Are there other example of transnational administrative procedures where BC use cases could be developed? In my article I examine what are the opportunities and challenges of using the DLT /BC in the context of official recognition of foreign diplomas (diplomas earned abroad) by governments. Now I wish to ask the views on that from the RG community of researchers.
Relevant answer
Answer
Many thanks Jaroslav Dvorak, I also came to the conclusion that there is a reluctance by states to adopt new technologies in their procedures for official recognition of foreign qualifications. But isn't that odd at a time where all governments are talking about digitization of their services to citizens and e-administration? I think it is worth deepening where the obstacles are: on the administration side or on the technology developer side?
Best regards
Christian
  • asked a question related to Recognition
Question
4 answers
Hi all.
I have a set of coordinates of bounding boxes for AVA Dataset. This dataset is used in action recognition. My goal is to get the areas around the people (people are inside boxes) and do saliency on those areas. How can I obtain those regions? When the person boxes are cropped how can I get the regions remained?
Relevant answer
Answer
Dear Sahar Pordeli Behrouz,
This link may answer your question,
Regards,
  • asked a question related to Recognition
Question
12 answers
Natural language understanding
Relevant answer
Answer
A Unified Architecture for Semantic Role Labeling and Relation Classification Jiang Guo, Wanxiang Che, Haifeng Wang , Ting Liu and Jun Xu
  • asked a question related to Recognition
Question
5 answers
Hello everyone, im currently working on a recognition system. I have a folder full of images and I need to convert the whole folder to .h5 file. this will help me with the next step in recognition system. How do I create one to proceed?
Relevant answer
Answer
How to create a HDF5 file from a directory of training images and a directory of validation images
  • asked a question related to Recognition
Question
4 answers
Hello everyone,i am young researcher,looking for MATLAB code on emotion recognition using EEG SIGNAL(SEED DATASET)?
Thank you
Relevant answer
Answer
I think the best solution is to search for any similar project, even if it uses a different feature, and then understand the project and write your own code related to your data.
I think any other project even in ECG or EMG will be fair enough to get the concept of classification and digital signal analysis.
I recommend you this link from the Matlab website that explain the basic concepts that you need to analyze any biosignal information.
I hope this answer can help you
  • asked a question related to Recognition
Question
5 answers
Hello, everyone!
I am a bachelor of software engineering (undergraduate) and I am currently conducting a research on the emotion recognition, and I need some help with the starting material. I found a lot of articles on that topic. But still I am just starting studying and I want to get some recommendations on books, articles or/and courses to create emotion recognition software program.
Also I am interested in physiological aspects of emotion recognition, so I will be grateful for some physiology materials too.
Relevant answer
Answer
I recommend doing a literature review on two specific databases SCOPUS and WEB OF SCIENCE. There you will find a lot of research directed to your area of knowledge regarding the Psychology of Emotions.
  • asked a question related to Recognition
Question
5 answers
Hi, is anyone familiar with a model/theory of memory recall and recognition in math education? Thank you
Relevant answer
Answer
For a start, try George Mandler's 1980 article in Psychological Review entitled Recognizing: The Judgment of Previous Occurrence.
  • asked a question related to Recognition
Question
4 answers
I have attached the files with this post to explain me how to remove rotation and center the origin at the body's local coordinate. This is just one sample and I have 1000s of files like this. I am working on a project based on affect recognition using autoencoders. I have been looking for sources to normalize the 3d data to make it suitable for training, but still I am struggling with removal of rotation and centring the origin. Please also refer to video to have an idea about the data.
Relevant answer
Answer
Khristian Edi Nugroho Soebandrija what it means to refer to baseline?
  • asked a question related to Recognition
Question
1 answer
What are the next steps in speaker identification after we extract the MFCC features? Thank you so much.
  • asked a question related to Recognition
Question
3 answers
I found that "open face" software works with linux operating system. Since I do not have linux, I would like to know if there is any other software similar to "open face" (even not free) that can be used for the recognition of emotions related to facial expressions.
Thank you!
Relevant answer
Answer
Veronica Maglieri another possibility is to use a virtual machine with a unix-like if you prefer. By the way, did you even hear about OpenCV? https://docs.opencv.org/master/d6/d00/tutorial_py_root.html
  • asked a question related to Recognition
Question
5 answers
My plasmid (pJET1.2) contains an insert of around 2969bp. The total size of plasmid 5942 bp. I am cutting it with AhdI, SgrAI and BssHII. Desired bands are of 2969, 1632 and 1432bp. I want to clone 2968bp gene. But I was getting bands of 3000,1550, 1350bp and also very light band of 625bp. when I gel-purified 3000bp band and run on gel then it was showing an extra band of 2700bp. Somehow I was able to clone this 3000bp but colony PCR shows a band of 2700bp instead of 2969bp. Also, RE confirmation shows a band size of 2700bp instead of 2969bp. Please anyone help me to solve this problem as I have repeated the experiment with so many times but getting similar results. My insert contains secondary sites that are similar to the recognition site of SgrAI but not exactly similar.
Relevant answer
Answer
As Reza said, the mobility difference between 2700 and 2969 may be hard to distinguish on an Agarose gel. It sounds like the primers that you are using for colony PCR are both on your vector; without spending much money, you could order a primer that binds to your insert and perform a colony PCR using a forward primer on your insert and a reverse primer on your vector-- This would at least answer if the 2700 band is your insert or not, which may assist your troubleshooting. If it is indeed the correct band and you don't have the resources to redesign this study, you could try to shorten your restriction enzyme digestion time or see if the manufacturer of your SgrAI and BssHII enzymes has a recommendation for buffer additives or alternative buffer formulations that reduce star activity.
  • asked a question related to Recognition
Question
3 answers
We launched a project on the recognition of haematoma in the CT of the human brain. This project was initiated on the basis of the Polenov Russian Research Institute of Neurosurgery.
At the moment, we have some practical outcomes which were discussed with the practitioners of the Polenov Institute.
We are interested in collaborators/colleagues from around the world for the discussing our outcomes.
Relevant answer
Answer
Thank you very much for the initiative! it will be a great work. our team will be happy to collaborate with your work.
Regards!
  • asked a question related to Recognition
Question
3 answers
Hybride intelligent system is good techniques used for solving diffrent optomization problems. Like classification, recognition ....etc
Deep learning can be used with and hybrid with other machine learning techniques. How we can optomized deep learning method by combining with swarm or fuzzy?
Relevant answer
Answer
This can happen by creating decentralised AI working model. In that decentralised model, swarm intelligence can be used for automating the resources allocation and better decisions making process. And the deep learning can be used for data or resources processing.
  • asked a question related to Recognition
Question
6 answers
Dear researchers i am working on content based video retrieval, i need popular datasets for evaluation of my method. I have downloaded UCF101,HMDB datasets but those datasets has very small clips specially collected for action recognition.
Relevant answer
  • asked a question related to Recognition
Question
5 answers
Using signal detection theory, we have calculated d prime for our research looking at false memories for faces.
We have found a significant correlation between facial recognition ability and d' for faces. Does that mean that the higher the d' prime score, the better you are at discriminating between hits and false alarms? Thus, your false memories for faces are lower?
Relevant answer
Answer
In answer to your second question about differences between d'=0, d'=1, d'=2, etc., I would give a somewhat more detailed answer. The logical range of d' is from 0 (chance) to 4.65 (based on Hit rate = 0.99, FA rate = 0.01). (Anything higher than d' = 4.65 will be getting awfully close to Hit = 1, and FA = 0, either of which give you "undefined" numbers [infinity].)
Importantly, because d' is a distance measure, and therefore has a true 0 value (no sensitivity at all), it is on a ratio scale. So, a d' = 2 is exactly twice as sensitive as d' = 1, and d' = 4 is twice as sensitive as d' = 2, etc. In this way, d' is like weight, and length, which are also ratio scales. Therefore, to take a more interesting example, since d' = 4.65 * 1/10 = 0.465, a d' of 0.465 is 10% of the way from chance (0) to maximum (4.65). That tells us that d' = 0.465 is quite poor. On the other hand, d' = 4.65 * 8/10 = 3.72, so a d' of 3.72 is 80% of the way from chance to maximum, which is really good (and 8x better than d' = 0.465).
  • asked a question related to Recognition
Question
4 answers
Emotion recognition is an upcoming research area with strong applications in the next-generation communications using digital modes. To develop robust algorithms for this, we are looking for some good datasets.
Relevant answer
Answer
Thank you Muhammad Ali, Omar Bouhamed and Moreno Colombo for sharing these.
  • asked a question related to Recognition
Question
4 answers
Hi,
I am interested in non-biological recognition elements for bacterial sensors. I am aware of molecularly imprinted polymers but I do not seem to find many other alternatives.
I know that many NPs and polymers are used for fluorescence-based assays but I am particularly curious about sensors, i.e. as alternatives to biosensors.
Any ideas and suggestions would be highly appreciated!
Relevant answer
Answer
Dear Karen Butina, the following documents may bring good insight into the scope of your question. My Regards
  • asked a question related to Recognition
Question
10 answers
I am graduate from other major, and now working in computer science and technology, my roomates are mostly researching on image recognition. I want to switch to it and learn how to cooperate with them. But I am still out of way.
Please help me how to gain the access to image recognition. I am skilled at nature inspired algorithms now. I have read some papers on image segmation but I don't know what to do. If you are skilled, could you show me a way?
Relevant answer
Answer
In my experience I suggest you to learn different segmentation in image processing and I recommend you using the famous library is OpenCV that is very powerful in computer science for reconnaissance of object or detection, we can fin you the example in websites of OpenCv, this example can help you to understand the implementation various segmentation.
I hope that be Claire for you@Zheng-Ming Gao
  • asked a question related to Recognition
Question
4 answers
Named entity recognition (NER) is task that mark tags of the input text sequence.
BERT-CRF is a good NER model.
I want to find a better NER model. Or I want to improve the BERT-CRF model.
What can I do?
Relevant answer
Answer
Hi Tong Guo , you can also try the following:
1.Deep Pavlov
3. Flair ( provides ensemble modeling feature too)
I would also encourage you to try multiple models as it depends from task to task.
One model might have faster inference
other may give near-perfect result for particular tags such as name or location.
So best model would be that does your task done more efficiently and may have some tradeoffs.
This one is a good paper I found on the subject https://arxiv.org/pdf/1812.09449.pdf
  • asked a question related to Recognition
Question
9 answers
Hello everyone, and I hope you are doing well.
I am working on doing the assessment of some rehabilitation exercises using machine learning algorithms.
For now, I am only trying to do real-time human activity recognition, I am collecting data from an IMU sensor (Shimmer device) and I want for example first to train a model offline on different activities for example sit and stand then to use that trained model for real-time recognition: I mean wearing the sensor and visualizing which activity I am performing on my pc using Python.
Could you please give me some information on how to proceed, which packages should I use, and the steps to follow.
If you have any code source example then please share it with me.
I have looked in different forums but couldn't find a relevant solution.
Thank you
Relevant answer
Answer
Arnaud Dion Aparna Sathya Murthy thank you for your Helpful answer, I will take a look
  • asked a question related to Recognition
Question
9 answers
I am doing my dissertation about brand recognition. I have 5 questions with 7-point Likert Scales that measure brand recognition. I want to combine those variables into one in order to conclude if the brand has recognition or not. How can I combine the variables? Will I go to Transform - Compute Variables in SPSS and will I combine them by the median? And after creating a new variable can I do the descriptive analysis? (like calculating the frequencies and the median)
Relevant answer
Answer
the seven items are not variables but they are indicators about brand recognition. If your data are normally distributed, calculated the mean and the standard deviation for each of the seven items, then calculate the overall mean and sd. If the data are skewed ly distributed, find the median for each of the seven items and calculate the overall median. In calculating the overall mean in SPSS, go to transform then go to compute variable, when you are here add all the responses of your total respondents divided by the number of respondents you have in your study. The result is now your overall mean. For overall median, combine all ratings of theyour respondents then determine the median using descriptive statistics.
  • asked a question related to Recognition
Question
4 answers
I am new to Machine Learning and I am currently doing research on speech emotion recognition using deep learning. I found out that recent literatures were using mostly CNN and there are only few literatures found for SER using RNN. I also found out that most approaches used MFCCs.
My questions are:
- Is it true that CNN has been proved to outperform RNN in SER?
- If yes, what are the limitations that RNN have compared with CNN?
- Also, what are the limitations of the existing CNN approaches in SER?
- Why MFCC is used the most in SER? Does MFCC have any limitations?
Any help or guidance would be appreciated.
Relevant answer
Answer
Answers to some of your queries are as follows:
- It depends on network configurations, the way one creates training examples & datasets. A lack of systematic benchmarking of existing methods however creates confusion. There are several studies which shows that LSTM outperforms CNN for speech emotion recognition. Specially LSTM with attention mechanism helps to boost emotion recognition performance. Some other studies report the opposite and CNN seems to be a better choice. You can do a quick literature survey on latest papers published in last couple of INTERSPEECH, ICASSP & ASRU.
- MFCC is the default choice for most speech processing tasks including speech emotion recognition. However, MFCC is not the optimal one as it lacks prosody information, long-term information. That's why often MFCC is augmented with pitch (to be more specific log F0) and/or shifted delta coefficients. These additional information help to boost the emotion recognition performance. MFCC lacks phase information but the role of phase in emotion recognition performance is not much investigated. The parameters for MFCC computation such as the number of filters, the frequency scale are chosen experimentally and they are dependent on the dataset and the backend classifier.
  • asked a question related to Recognition
Question
3 answers
Looking for information on training emotions using VR technology with psychology practices.
Any links/help appreciated
Relevant answer
Nice Dear Joris Heyse
  • asked a question related to Recognition
Question
3 answers
Hi there,
Recently I came across Computer Expression Recognition Toolbox (CERT) for facial expression analysis but it seems they no longer support anymore.
I am looking for an opensource tool that I can detect eyebrow-raising/ eyelid tightening/ blink data. Can you please recommend any?
Relevant answer
A