We provide the IAL model code as supplementary material and publish the DiffBR model code after acceptance of the paper.
- Recommended Environment: python 3.8.8, Cuda11.6, PyTorch 1.12.1(The PyTorch version should be at least >= 1.11.)
- Install dependencies:
pip install -r requirements.txt
- Install NMS:
cd ./libs/utils; python setup.py install --user; cd ../..
Download Features and Annotations
Dataset | Feature Encoder | Link |
---|---|---|
THUMOS14 | I3D | thumos_i3d |
THUMOS14 | VideoMAE | thumos_videomae |
ActivityNet | I3D | anet_i3d |
ActivityNet | TSP | anet_tsp |
ActivityNet | VideoMAE | anet_videomae |
Epic-Kitchen | SlowFast | epic_kitchen |
Unpack Features and Annotations
- Unpack the file under
./data
- The folder structure should look like
DAD-TAD/
├── data
│ ├── anet_1.3
│ │ ├── annotations
│ │ ├── i3d_features
│ │ ├── tsp_features
│ │ └── anet_mae_hugek700
│ ├── epic_kitchens
│ │ ├── annotations
│ │ ├── features
│ └── thumos
│ ├── annotations
│ ├── i3d_features
│ ├── th14_mae_g_16_4
├── libs
├── tools
└── ...
-
We have provided a script list that allows you to replicate our results with just a single click. Further details can be found in
./tools/run_all_exps.sh
. -
Our experiments were conducted exclusively on a single NVIDIA GeForce GTX 1080 Ti. It is noted that variations in the graphics card model may lead to slight discrepancies in replicating the results.
-
[Optional] Monitor the training using TensorBoard. Example:
tensorboard --logdir=./ckpt/thumos_i3d_final/logs