search

facebookarchive / C3D

C3D is a modified version of BVLC caffe to support 3D ConvNets.
Other
1.17k stars 507 forks source link

Features are not being extracted for all the frames #434

Open kjsr7 opened 5 years ago

kjsr7 commented 5 years ago

I have build the c3d-v1.0 model successfully and tried to extract the features for my .mp4 video for every 16 frames. I have generated the 'input_list_video.txt' and 'output_list_video_prefix.txt'. The features are getting extracted successfully. The problem is for all the videos only 50 output files are being generated which span only some frames, but not the remaining frames. input_list_video.txt output_list_video_prefix.txt

I have even tried to modify the mini_batch_size from 50 to 60, but no use. Could you please help me in extracting the features for all my frames?

kjsr7 commented 5 years ago

I think I need to change the number of mini batches in such a way that (batchsize * number of mini batches >= number of frames) in my video. @dutran, could you please confirm?

Shiny128 commented 5 years ago

Hey @kjsr7 , How did you extract features for new videos? Which file did you use for execution?

kjsr7 commented 5 years ago

@Shiny128, Use this for extracting features for new videos.

hemaprabha commented 5 years ago

@kjsr7 Hi can u please tell me the steps how to run this model from the scratch .As I'm new to this model.

YuliaNovskaya commented 5 years ago

@kjsr7 I have the same problem. Could you please tell me, have you solved this? Thank you in advance