saliency VS darkflow

Found relevant code at https://github.com/PAIR-code/saliency + all code implementations here

I’m trying to visualize model layer outputs using the saliency core package package on a simple conv net. This requires me to compute the gradients of the model output layer and intermediate convolutional layer output w.r.t the input. I’ve attempted to do this in the last code block, but I run into the error

[2]: https://github.com/PAIR-code/saliency

https://github.com/amusi/awesome-object-detection https://mmdetection.readthedocs.io/en/latest/index.html https://github.com/thtrieu/darkflow https://github.com/OlafenwaMoses/ImageAI https://github.com/dmlc/gluon-cv https://github.com/aim-uofa/AdelaiDet/ https://github.com/aim-uofa/AdelaiDet/blob/master/configs/FCOS-Detection/README.md https://github.com/wizyoung/YOLOv3_TensorFlow

You could already tell from my earlier comment, but I don't think Python is a good language, for anything really. It is used for AI so that today's hardware performs like hardware we had twenty years ago: https://github.com/thtrieu/darkflow/issues/904 Python code has slow execution speed. It is not always the Selenium which is slow. sometimes we need to look at code we are using. And Python is always the slowest programming language out there in terms of performance. It's also just a stupid language that annoys me and it's the same with other people: https://medium.com/nerd-for-tech/python-is-a-bad-programming-language-2ab73b0bda5 With Python, I find it a bit too easy to write sloppy code. Haskell on the other hand really forces you to break the problem done and abstract out reusable code. That's not to say that you can't write nice code in Python, just that Haskell doesn't let you get away with doing a lot of stupid stuff. This is equally important for Selenium.