This week we mainly planned on how to reform our project in order to work on it remotely. I was formerly responsible for designing and building the physical product. We have decided that is not viable to pursue this goal. Additionally, we are no longer using a Raspberry PI to run the project code. We’re aiming to have each of our emotion recognition modules done, and if time allows, full integration on a server. Instead, Read more…
This week, the team -Refocused our project to transition into working remotely -Set up team zoom meetings -Changed final product from physical device to server-hosted application -Added new part to our project to make up for loss of physical product To do: -Work on individual components -Start planning how to combine different analyses -Set up cloud Our Statement of Work here: F5_SOW
This week was mainly focused on planning how to move forward with our project, what to keep and what to change. More details about this can be found below in my statement of work, but the main part is that I will have a new task in the project, speech tone analysis. I gathered up the datasets and researched previous projects that used this (unfortunately there aren’t many), downloaded software (Praat) to analyze voice waveforms. Read more…
Completed: Refocus SOW Setting up Zoom Migrating FER algorithm to home desktop computer Setting up PyCharm, Tensorflow, Keras, SkLearn, OpenCV Downloading datasets Reworking code to work in new environment To-do: Continue using more of AffectNet database Find more ways to improve accuracy Tweaking SVM parameters like bias, loss, etc Perform cross-validation on AffectNet Test if algorithm works faster in real-time on desktop computer Tweak number of times algorithm is run per second to meet requirements
This week I mostly worked on our design report. I also worked on setting up google speech to text and using sklearn’s word vectorizer. With two emotions, I was able to achieve around 80% accuracy with the word vectorizer. Going forward, I’m planning to work on the LSTM-CNN for textual emotion analysis. From my research online, using one LSTM layer and one or two CNN layers produces optimal results:. I do not think I’ll make Read more…
This week I worked on the UI of the web application. Updated look here: I also added the six emotion categories (sad, happy, angry, neutral, surprise, disgust) properly into the database; emotion categories can now be linked to entries. More things I need to do is to make the numbers of emotions fixed and types unique (so that someone can’t add more categories somehow) Did not make as much progress on the design as I Read more…
This week we finished our design report and collected most of our parts for our hardware. We still need a screen and microphone. We need to set up an AWS server since we also got our AWS credits. We are still working on our individual components: training the sentiment analysis and facial emotion recognition and also working on the web app. Next week is spring break so we will not really be working as much Read more…
Completed tasks: Test preliminary trained SVM Have SVM prediction in real-time Include AffectNet dataset in SVM training Design report To-do: Train with more AffectNet images Figure out how to output with probabilities for each emotion instead of single output Improve accuracy Set up AWS and train on server
This week I worked on the design presentation and the web app; working on the presentation helped me plan out the web app more concretely; for example the models and url paths, included in the presentation and also added here: My goal for this coming week is to finish implementing all the models (I am going to add user authentication as well, for privacy reasons regarding the entries; it will be better and easier to Read more…
Completed tasks: Automated I/O of CK+ dataset Localized pre-processed VGG19 features of CK+ dataset for faster training Preliminary training of SVM with 327 labeled pictures from CK+ dataset Working on evaluating trained SVM Implemented real-time prediction with camera input To-do: Evaluate performance of preliminary trained SVM Speed up real-time SVM prediction Improve accuracy of SVM prediction Include AffectNet dataset in SVM training Design report