• This only test ECO-Full model for dataset UCF101
• You can try it on AIStudio project https://aistudio.baidu.com/aistudio/projectdetail/698987

• Python 3.7
• PaddlePaddle 1.8.0

git clone https://github.com/Nullius-2020/ECO-pp

python gen_dataset_lists.py <ucf101> <dataset_frames_root_path>

e.g. python gen_dataset_lists.py ucf101 ~/dataset/ucf101/

The dataset should be organized as:
<dataset_frames_root_path>/<video_name>/<frame_images>

For finetuning on UCF101 use the following command:

sh run_demo_ECO_Full.sh local

### NOTE
* If you want to train your model from scratch change the config as following:
```bash
    --pretrained_parts scratch

• configurations explained in "opts.py"

• This project originated from the course https://aistudio.baidu.com/aistudio/education/group/info/1340

If you use this code or ideas from the paper for your research, please cite this paper:

@inproceedings{ECO_eccv18,
author={Mohammadreza Zolfaghari and
               Kamaljeet Singh and
               Thomas Brox},
title={{ECO:} Efficient Convolutional Network for Online Video Understanding},	       
booktitle={ECCV},
year={2018}
}