search

VisionLearningGroup / R-C3D

code for R-C3D
MIT License
255 stars 95 forks source link

about the ucf101.caffemodel #15

Open applesleam opened 6 years ago

applesleam commented 6 years ago

Hi @huijuan88 Thank you for updating the code of thumos14 and charades! I want to run the training on thumos14, however I found the ucf101 fine-tuned model provided by C3D v1.0 is not compatible with your code. And I could not find the newest version (C3D-v1.1) caffe model fine-tuned on ucf101. Could you plz tell me how can I find the v1.1 model or provide us your pretrained model for thumos14? Thanks a lot!

huijuan88 commented 6 years ago

I convert the ucf101 weights from old C3D. I don't have corresponding model in C3D-v1.1.

On Mar 25, 2018, at 04:42, Qing Wang notifications@github.com<mailto:notifications@github.com> wrote:

Hi @huijuan88https://github.com/huijuan88 Thank you for updating the code of thumos14 and charades! I want to run the training on thumos14, however I found the ucf101 fine-tuned model provided by C3D v1.0 is not compatible with your code. And I could not find the newest version (C3D-v1.1) caffe model fine-tuned on ucf101. Could you plz tell me how can I find the v1.1 model or provide us your pretrained model for thumos14? Thanks a lot!

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHubhttps://github.com/VisionLearningGroup/R-C3D/issues/15, or mute the threadhttps://github.com/notifications/unsubscribe-auth/AFOa__kOuep3Zbk86p6nzoPni6bt4BBPks5th1hggaJpZM4S6D1e.

applesleam commented 6 years ago

Thanks for your reply! I'll try it~

applesleam commented 6 years ago

@huijuan88 Sorry to bother again~ I thought your model was trained based on C3D-V1.1, however I found that your caffe3d could not read the v1.1 prototxt file, and when I load the v1.1 weights into your C3D model here comes the problem with shape mismatch. Like this: F0404 09:43:43.854476 3739 net.cpp:774] Cannot copy param 1 weights from layer 'conv1a'; shape mismatch. Source param shape is 1 1 1 1 64 (64); target param shape is 64 (64). To learn this layer's parameters from scratch rather than copying from a saved net, rename the layer. I also have no idea how to convert the ucf101 weights from old C3D into your weight format. Could you plz give me any hint or reference?

huijuan88 commented 6 years ago

I didn't use C3D-V1.1. Thus it can't be directly loaded.

Read weights, assign, and save.