search

chuckcho / video-caffe

Video-friendly caffe -- comes with the most recent version of Caffe (as of Jan 2019), a video reader, 3D(ND) pooling layer, and an example training script for C3D network and UCF-101 data
Other
175 stars 93 forks source link

Hello， I wonder to know why this training accuracy is about 0.5, but the accuracy in the C3D paper is about 0.8? thanks! #101

Closed ARQlalala closed 6 years ago

ARQlalala commented 6 years ago

Please use the caffe-users list for usage, installation, or modeling questions, or other requests for help. Do not post such requests to Issues. Doing so interferes with the development of Caffe.

Please read the guidelines for contributing before submitting this issue.

Issue summary

Steps to reproduce

If you are having difficulty building Caffe or training a model, please ask the caffe-users mailing list. If you are reporting a build error that seems to be due to a bug in Caffe, please attach your build configuration (either Makefile.config or CMakeCache.txt) and the output of the make (or cmake) command.

Your system configuration

Operating system: Compiler: CUDA version (if applicable): CUDNN version (if applicable): BLAS: Python or MATLAB version (for pycaffe and matcaffe respectively):

chuckcho commented 6 years ago

@ARQlalala 50% accuracy you're getting is when you train from a scratch, while 80% accuracy in the paper is based on fine-tuning from a pre-trained model (Sport1m).

chuckcho commented 6 years ago

See figure 2 in the original paper which is comparable to ~50% number you're getting.