add usage and to-do list

README.md

@@ -24,6 +24,24 @@ Some key dependencies are listed below, while others are given in [`requirements
 - PyTorch >= 1.3, and a corresponding version of torchvision
 - ffmpeg (used in data preparation)
 
+## Usage
+Default values for arguments `nproc_per_node`, `backend` and `master_port` are `8`, `nccl` and `31114` respectively.
+
+```
+python main.py --config CONFIG_FILE [--nproc_per_node N_PROCESSES] [--backend BACKEND] [--master_addr MASTER_ADDR] [--master_port MASTER_PORT]
+```
+
+### Running with Multiple Machines
+In this case, the `master_addr` argument must be provided. Moreover, arguments `nnodes` and `node_rank` can be additionally specified (similar to `torch.distributed.launch`), otherwise the program will try to obtain their values from environment variables. See [`distributed_utils.py`](https://github.com/Siyu-C/ACAR-Net/blob/master/distributed_utils.py) for details.
+
+## To-do List
+- Data preparation for AVA dataset
+- Pre-trained models
+- Model zoo
+- More advanced backbone
+- Data preparation for Kinetics dataset, and training on AVA-Kinetics
+- Implementation for ACFB
+
 ## License
 ACAR-Net is released under the [Apache 2.0 license](https://github.com/Siyu-C/ACAR-Net/blob/master/LICENSE).
 

main.py

@@ -8,6 +8,7 @@
 import time
 from easydict import EasyDict
 import yaml
+from tensorboardX import SummaryWriter
 
 import torch
 import torch.optim as optim
@@ -16,15 +17,12 @@
 
 from calc_mAP import run_evaluation
 from datasets import ava, spatial_transforms, temporal_transforms
+from distributed_utils import init_distributed
 import losses
 from models import AVA_model
 from scheduler import get_scheduler
 from utils import *
 
-from distributed_utils import init_distributed
-
-from tensorboardX import SummaryWriter
-
 
 def main(local_rank, args):
     '''dist init'''
@@ -437,7 +435,7 @@ def val_epoch(epoch, data_loader, model, criterion, act_func,
 
 
 if __name__ == '__main__':
-    parser = argparse.ArgumentParser(description='PyTorch AVA Training')
+    parser = argparse.ArgumentParser(description='PyTorch AVA Training and Evaluation')
     parser.add_argument('--config', type=str, required=True)
     parser.add_argument('--nproc_per_node', type=int, default=8)
     parser.add_argument('--backend', type=str, default='nccl')

models/backbones/slowfast.py

@@ -1,13 +1,7 @@
 """
-Reference: [SlowFast Networks for Video Recognition](https://arxiv.org/abs/1812.03982).
-Adapted code from:
-    @inproceedings{hara3dcnns,
-      author={Kensho Hara and Hirokatsu Kataoka and Yutaka Satoh},
-      title={Can Spatiotemporal 3D CNNs Retrace the History of 2D CNNs and ImageNet?},
-      booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
-      pages={6546--6555},
-      year={2018},
-    }.
+References:
+[SlowFast Networks for Video Recognition](https://arxiv.org/abs/1812.03982),
+[PySlowFast](https://github.com/facebookresearch/slowfast).
 """
 
 import torch