Refine Align-DETR project (#284)

* refine align-detr

README.md

@@ -80,6 +80,7 @@ The repo name detrex has several interpretations:
 ## What's New
 v0.5.0 was released on 16/07/2023:
 - Support [SQR-DETR (CVPR 2023)](https://github.com/IDEA-Research/detrex/tree/main/projects/sqr_detr), credits to [Fangyi Chen](https://github.com/Fangyi-Chen)
+- Support [Align-DETR (ArXiv 2023)](./projects/align_detr/), credits to [Zhi Cai](https://github.com/FelixCaae)
 - Support [EVA-01](https://github.com/baaivision/EVA/tree/master/EVA-01) and [EVA-02](https://github.com/baaivision/EVA/tree/master/EVA-02) backbones, please check [DINO-EVA](./projects/dino_eva/) for more benchmarking results.
 
 Please see [changelog.md](./changlog.md) for details and release history.
@@ -129,6 +130,7 @@ Results and models are available in [model zoo](https://detrex.readthedocs.io/en
 - [x] [MaskDINO (CVPR'2023)](./projects/maskdino/)
 - [x] [CO-MOT (ArXiv'2023)](./projects/co_mot/)
 - [x] [SQR-DETR (CVPR'2023)](./projects/sqr_detr/)
+- [x] [Align-DETR (ArXiv'2023)](./projects/align_detr/)
 - [x] [EVA-01 (CVPR'2023 Highlight)](./projects/dino_eva/)
 - [x] [EVA-02 (ArXiv'2023)](./projects/dino_eva/)
 

projects/README.md

@@ -15,5 +15,6 @@ Here are projects that are built on detrex which show you use detrex as a librar
 - [NMS strikes back (ArXiv'2022)](./deta/)
 - [CO-MOT: Bridging the Gap Between End-to-end and Non-End-to-end Multi-Object Tracking (ArXiv'2023)](./co_mot/)
 - [Enhanced Training of Query-Based Object Detection via Selective Query Recollection (CVPR'2023)](./sqr_detr/)
+- [Align-DETR: Improving DETR with Simple IoU-aware BCE loss (ArXiv'2023)](./align_detr/)
 - [EVA-01: Exploring the Limits of Masked Visual Representation Learning at Scale (CVPR'2023 Highlight)](./dino_eva/)
 - [EVA-02: A Visual Representation for Neon Genesis (ArXiv'2023)](./dino_eva/)

projects/aligndetr/README.md → projects/align_detr/README.md

@@ -1,20 +1,27 @@
 # AlignDETR
-## Introduction
+
+Zhi Cai, Songtao Liu, Guodong Wang, Zheng Ge, Xiangyu Zhang, Di Huang
+
+[[`arXiv`](https://arxiv.org/abs/2304.07527)] [[`BibTex`](#citation)] [[`Official Implementation`](https://github.com/FelixCaae/AlignDETR)]
+
+<img src="https://github.com/FelixCaae/AlignDETR/blob/main/assets/aligndetr.png?raw=true" >
+
 AlignDETR is a variant of DETR(DEtection with Transformer), with a simple IoU-Aware BCE loss and better performance! It aims to solve the issue of misalignment problem spotted in DETR's output.
-For more details, please refer to our [paper on Arxiv](https://arxiv.org/abs/2304.07527). 
-<img src="assets/aligndetr.png" >
 
 ## Installation 
 See [installation instructions](INSTALL.md) 
+
 ## Usage
 Train Example
 ```bash
 python tools/train_net.py --config-file  aligndetr/aligndetr_k=2_r50_4scale_12ep.py --num-gpus 8
 ```
+
 Evaluation Example
 ```bash
 python tools/train_net.py --config-file  aligndetr/aligndetr_k=2_r50_4scale_12ep.py --num-gpus 8 --eval train.init_checkpoint=/path/to/checkpoint
 ```
+
 ## Model Zoo
 
 |Model|AP|AP50|AP75|APs|APm|APl|weights|
@@ -25,6 +32,7 @@ python tools/train_net.py --config-file  aligndetr/aligndetr_k=2_r50_4scale_12ep
 ## Acknowlegements
 
 Our code is based on [detrex](https://github.com/IDEA-Research/detrex) and [detectron2](https://github.com/facebookresearch/detectron2).
+
 ## Citation
 
 If you are interested in our work and use our method in your research, please cite