C3D-Tensorflow-slim

A simple Tensorflow code for C3D

Reqirements

Tensorflow(>=1.4) Python(>=2.7)

UCF101. You need to place it in the root directory of the workspace.

sudo ./convert_video_to_images.sh UCF101/ 5

to convert videos into images(5FPS per-second).

./convert_images_to_list.sh UCF101/ 4

to obtain train and test sets.(3/4 train; 1/4 test)

python train.py

for training.

python test.py

for testing.

Radom choose 16 frames as a clip to represent a video for training and testing.

0.72 on the test set after 36 epoches.

The files convert_video_to_images.sh, convert_images_to_list.sh and crop_mean.npy are copied from https://github.com/hx173149/C3D-tensorflow