Week 8 : Coding Period
This week I focused on the Chest-Xray14 dataset and its available models. The benchmark model for this dataset is the Chexnet model. This was availabale on Github in 2 formats - Pytorch and Tensorflow. So I downloaded the Pytorch model and set up a pipeline to convert this model from Pytorch to ONNX to Tensorflow. Part 1 of converting the Pytorch model to ONNX was implemented successfully inspite of a lot of bugs (because DataParallel is not supported by ONNX). Now the part of converting the ONNX model to Tensorflow is generating a lot of errors. Errors that sometimes have no solutions available on the internet! This section will require me to look into it thoroughly.
During this course, I also searched for available Keras models of Chexnet. The one that I found did not have any concrete results. Because the creator did not add a threshold/classifier. The model simply outputs scores per class. Another fishy aspect of this model is that it is only 28MB in size. How can such a heavy, un-compressed model have such a low size? It would either be wise to add additional layers to this model and then use it, or create my own version of chexnet in Tensorflow and train it (with concrete results).
I have also been writing docstrings for the python scripts that I created for training and compressing models based on the RSNA Pneumonia Detection Dataset. I will commit this part of my code shortly.
During the first few days of this week, I finished running the Int8 evaluation scripts (that took anywhere between 12 to 24 hours to run) for my quantized pruned models. I presented these results to my mentor and I also observed that some of these models gave results which were same to the purely quantized models.
Since I have run into these many roadblocks this week, I will discuss with my mentor to decide my goals for this week as well as the upcoming month.
Happy coding!
During this course, I also searched for available Keras models of Chexnet. The one that I found did not have any concrete results. Because the creator did not add a threshold/classifier. The model simply outputs scores per class. Another fishy aspect of this model is that it is only 28MB in size. How can such a heavy, un-compressed model have such a low size? It would either be wise to add additional layers to this model and then use it, or create my own version of chexnet in Tensorflow and train it (with concrete results).
I have also been writing docstrings for the python scripts that I created for training and compressing models based on the RSNA Pneumonia Detection Dataset. I will commit this part of my code shortly.
During the first few days of this week, I finished running the Int8 evaluation scripts (that took anywhere between 12 to 24 hours to run) for my quantized pruned models. I presented these results to my mentor and I also observed that some of these models gave results which were same to the purely quantized models.
Since I have run into these many roadblocks this week, I will discuss with my mentor to decide my goals for this week as well as the upcoming month.
Happy coding!
Comments
Post a Comment