Pascal experiments code or models?

[jason718]

Do you use some pytorch version of pascal det., cls., and seg. code? Or do you convert your model into caffemodel?

[gidariss]

I convert the models on caffe format and use the publicly available caffe code. I uploaded the readme.md with link to a model converted to caffe format.

[jason718]

Can you share your converting code? Or are you using the something like pytorch-caffe-darknet-convert?

[polo5]

+1. It would also be nice for instance to have the configuration you used for classification scores (https://github.com/philkr/voc-classification). I can't reproduce the results in your paper with that repo.

[gidariss]

#jason718

I updated the github repo with the a script that converts a pytorch alexnet model to a caffe model. It is inside the directory "extras". The same directory also has a script that converts the caffe alexnet model to fully convolutional for the segmentation experiments. When I find time I will update the README with instructions for that part. However, it's pretty easy to use the conversion script: after having install caffe you run the command:

python extras/convert_alexnet_from_pytorch2caffe.py --src=[path to pytorch model] --dst=[path to caffe mode]

[iamgeo92]

hi @gidariss , Does the script also contains the code for rescaling of Kraehenbuehl et al, ICLR 2016.