Week 9 - Coding Period

-
Update on the project -

This week I tried converting the Chexnet Pytorch model to Onnx and then to Tensorflow again. As Onnx does not support DataParallel, I tried this conversion without using it. I wrote a regex to convert the state dictionary keys of the Pytorch model as the model was outdated and the keys needed to be renamed. I was able to convert the model to Onnx format. Inspite of these changes, I ran into a lot of errors while converting this new model to Tensorflow. So I dropped this model and began working on the purely Tensorflow model. This is not the standard model, but it gives considerably good results. I am currently converting the output format of this model to represent the actual class names instead of probability values and compressing these too.
 
I added modifications to the scripts to convert pruned models to quantized ones. I need to clean this code to specify the different input formats for models - hdf5 and h5 models. I have evaluated my models for size and accuracy. After completing these tasks, I want to evaluate them for latency, power consumption and computational cost. This may differ from device to device.

The second evaluation results came this week and my mentor has given me suggestions to modify my repository. I need to restructure my project and correct my errors. The scripts need to separated and I also need to make provisions for a few remaining edge cases.

I also need to structure my timeline to achieve substantial results by the end of this month.

Till then, Happy Coding!

Comments

Post a Comment

Popular posts from this blog

GSoC 2020 with LibreHealth : Final Report

-
Image
Low Powered Models for Disease Detection and Classification for Radiology Images Project Description - The aim of this project is to create Deep Learning models for detection and classification of radiology images. The models must be compressed such that they can be deployed to low powered devices like ARM devices, Android devices, etc. Compression techniques such as Quantization and Pruning can be used.     Mentors - Priyanshu Sinha Saptarshi Purkayastha Judy Gichoya Geeta Priya Padmanabhan Tech Stack -  Numpy Pandas PyDicom Tensorflow Tensorflow-Lite/ Tensorflow-Model-Optimization Docker Qemu Project Link - Click here Commits - Click here    Merge Requests - Click here Why to do this - There has been a lot of progress in developing Machine Learning models that predict the medical condition of a patient based upon specific inputs relevant to the diagnosis of that condition. However, these models have drawbacks while deployment in real-time on edge devices....
Read more

Week 1 : Acceptance and Community Bonding

-
Image
Introduction My name is Aishwarya Harpale and I am a Senior year Computer Engineering student from Pune Institute of Computer Technology, Pune. I was selected for GSoC 2020 to contribute to the Open Source Organization — LibreHealth. The project that I will be working on is “Low Powered Models for Disease Detection and Classification for Radiology Images”. I will explain a bit about my journey until now. Pre-GSoC Period After the organizations were announced, I began looking for projects that came under the category of Machine Learning. LibreHealth’s projects caught my eye and I felt that I could contribute to those projects using my knowledge. As students, we were asked to develop a Proof of Concept(PoC) so that the mentors could evaluate our coding skills. I developed an application for Classification and Localization of Chest-XRay images. This application had a front-end developed using Bootstrap and it was deployed on a Flask server. For further detai...
Read more

Week 2 : Coding Period

-
This week I continued training models on the dataset. I was running into OOM errors because the complete dataset could not be loaded into the memory all at once. I overcame this error by using mini-batching. This method loads a small set of data into the memory as an when required as per the training batch size. It reduces the overload on the memory as well. I trained DenseNet201 and InceptionV3 on the dataset. These models did not give sufficient accuracy which is why I will be experimenting with more hyper-parameters before proceeding to the next task of quantization. The current accuracy is around 85%. I want to stretch it to a scale above 95%. This will be my focus this week. Happy coding! :)
Read more