search

irhum / R2Plus1D-PyTorch

PyTorch implementation of the R2Plus1D convolution based ResNet architecture described in the paper "A Closer Look at Spatiotemporal Convolutions for Action Recognition"
MIT License
363 stars 62 forks source link

Does this project provide the pretrained model? #1

Open liu-zhy opened 6 years ago

liu-zhy commented 6 years ago

Please forgive me to new this issues. Is there providing the pretrained weights? Or we train it from scratch by ourselves? Could you mind share it to all of us？Becase somebody like me does not have so much gpus due to our limited fundings. Thanks for your generosity！

whai362 commented 6 years ago

So do I.

irhum commented 6 years ago

The best weights available are the ones found in the original implementation, but they're Caffe2 weights. Currently looking into how to import them into PyTorch, any assistance would be welcome

swathikirans commented 5 years ago

Have you succeeded in loading the caffe2 weights to the provided pytorch model?

BibratRanjan commented 5 years ago

I'am trying to load the pickle weights (r2.5d_d34_l32_ft_sports1m.pkl) using :

with open('r2.5d_d34_l32_ft_sports1m.pkl', 'rb') as fopen: blobs = pickle.load(fopen)

But I'am getting errors. I'am attempting to convert caffe2 weights to pytorch. Any Ideas what is going wrong?

BibratRanjan commented 5 years ago

I'am using python 3.6

Caoliangjie commented 5 years ago

Have you finished convert the pretrained model to pytorch？

daveboat commented 4 years ago

Late reply here, but if anyone still wants pretrained weights, torchvision has kinetics-pretrained models for three of the architectures in the R(2+1)D paper:

https://pytorch.org/docs/stable/torchvision/models.html#video-classification