model_interpretability_captum
- Using Captum library and interpreting the models to understand their decisions better.
- the folowing results are on CIFAR10 and most of the code is from the captum.ai site.
- Captum can be easily used to integrate with python but it requires high amount of RAM.
- I tried using it on a resnet18 with (224 & 128) resolution and my 16GB machine was not enough.
- Resolution of (60 * 60) does work though.
- You can try the code (Main-cats&dogs.ipynb) if you have enough RAM.
- I will be uloading results once I have access to a capable Machine.
- I've tested with PyTorch 1.5 but I couldn't run it on PyTorch 1.1.0(It doesn't support)
Installaion
conda install pytorch torchvision cpuonly -c pytorchconda install -c captum pytorch
Results
