Skip to content

Latest commit

 

History

History

object_detection

README.md

Applying TransXNet to Object Detection and Instance Segmentation

For details, please address "TransXNet: Learning Both Global and Local Dynamics with a Dual Dynamic Token Mixer for Visual Recognition".

1. Requirements

# Environments:
cuda==11.3
python==3.8.15
# Packages:
mmcv==1.7.1
mmdet==2.28.2
timm==0.6.12
torch==1.12.1
torchvision==0.13.1

2. Data Preparation

Prepare COCO 2017 according to the guidelines.

3. Main Results on COCO with Pretrained Models

Method Backbone Pretrain Lr schd Aug box AP mask AP Config Download
RetinaNet TransXNet-T ImageNet-1K 1x No 43.1 - config log & model
RetinaNet TransXNet-S ImageNet-1K 1x No 46.4 - config log & model
RetinaNet TransXNet-B ImageNet-1K 1x No 47.6 - config log & model
Mask R-CNN TransXNet-T ImageNet-1K 1x No 44.5 40.7 config log & model
Mask R-CNN TransXNet-S ImageNet-1K 1x No 47.7 43.1 config log & model
Mask R-CNN TransXNet-B ImageNet-1K 1x No 48.8 43.8 config log & model

4. Train

To train TransXNet-T + RetinaNet models on COCO train2017 with 8 gpus (single node), run:

bash dist_train.sh configs/retinanet_transx_t_fpn_1x_coco.py 8

To train TransXNet-T + Mask R-CNN models on COCO train2017 with 8 gpus (single node), run:

bash dist_train.sh configs/mask_rcnn_transx_t_fpn_1x_coco.py 8

5. Validation

To evaluate TransXNet-T + RetinaNet models on COCO val2017, run:

bash dist_test.sh configs/retinanet_transx_t_fpn_1x_coco.py /path/to/checkpoint_file 8 --out results.pkl --eval bbox

To evaluate TransXNet-T + Mask R-CNN models on COCO val2017, run:

bash dist_test.sh configs/mask_rcnn_transx_t_fpn_1x_coco.py /path/to/checkpoint_file 8 --out results.pkl --eval bbox segm

Citation

If you find this project useful for your research, please consider citing:

@article{lou2023transxnet,
  title={TransXNet: Learning Both Global and Local Dynamics with a Dual Dynamic Token Mixer for Visual Recognition},
  author={Lou, Meng and Zhou, Hong-Yu and Yang, Sibei and Yu, Yizhou},
  journal={arXiv preprint arXiv:2310.19380},
  year={2023}
}

Contact

If you have any questions, please feel free to create issues or contact me at [email protected].