Skip to content

Latest commit

 

History

History
47 lines (28 loc) · 2.8 KB

GET_STARTED.md

File metadata and controls

47 lines (28 loc) · 2.8 KB
Raw

Get Started

In this documentation, we will primarily focus on training and inference of our MOTIP model on the relevant MOT benchmarks.

Training

In the ./configs/ directory, we provide various configuration files. You can use different configuration files to train the MOTIP model with different settings on various datasets.

Default Scripts

All our training scripts follow the template script below. You'll need to fill the <placeholders> according to your requirements:

python -m torch.distributed.run --nproc_per_node=8 main.py --mode train --use-distributed True --use-wandb False --config-path <config file path> --data-root <DATADIR> --outputs-dir <outputs dir>

For example, you can the default model on DanceTrack as follows:

python -m torch.distributed.run --nproc_per_node=8 main.py --mode train --use-distributed True --use-wandb False --config-path ./configs/r50_deformable_detr_motip_dancetrack.yaml --data-root ./datasets/ --outputs-dir ./outputs/r50_deformable_detr_motip_dancetrack/

Using this script, you can achieve 66.2 ~ 67.6 HOTA on DanceTrack test set. This relatively high instability (~ 1.5) is also encountered in other work (e.g., OC-SORT, MOTRv2, MeMOTR). We suggest that part of the reason comes from the DanceTrack dataset itself, because the final performance on the MOT17 or SportsMOT test set will be more stable (~ 0.2 HOTA and ~ 0.5 HOTA).

Gradient Checkpoint

If your GPUs have less than 24GB CUDA memory, we offer the gradient checkpoint technology. You can set --detr-checkpoint-frames to 2 (< 16GB) or 1 (< 12GB) to reduce the CUDA memory requirements.

Inference

Submitting

To evaluate on most MOT benchmarks, you’ll need to submit tracking result files to the official server. You can obtain the tracking output using the following script:

python -m torch.distributed.run --nproc_per_node=<gpu num> main.py --mode submit --use-distributed True --use-wandb False --config-path <config file path> --data-root <DATADIR> --inference-model <checkpoint path> --outputs-dir <outputs dir> --inference-dataset <dataset name> --inference-split <dataset split>

For example, you can submit the model on DanceTrack test set as follows:

python -m torch.distributed.run --nproc_per_node=8 main.py --mode submit --use-distributed True --use-wandb False --config-path ./configs/r50_deformable_detr_motip_dancetrack.yaml --data-root ./datasets/ --inference-model ./outputs/r50_deformable_detr_motip_dancetrack.pth --outputs-dir ./outputs/dancetrack_trackers/ --inference-dataset DanceTrack --inference-split test

Evaluation

🚧 The evaluation scripts will be available in the future.