🧑‍🏫 Consistent-Teacher: Towards Reducing Inconsistent Pseudo-targets in Semi-supervised Object Detection 🧑‍🏫

This repository contains the offical implementation for our CVPR-2023 paper.

✨We are now able to train detector on 10% MS-COCO to 40 mAP✨

Consistent-Teacher: Towards Reducing Inconsistent Pseudo-targets in Semi-supervised Object Detection

[arxiv] [code] [project page]

Xinjiang Wang*, Xingyi Yang*, Shilong Zhang, Yijiang Li, Litong Feng, Shijie Fang, Chengqi Lyu, Kai Chen, Wayne Zhang

(*: Co-first Author)

• [x] Selected as Hightligh for CVPR2023🔥 (235/2360, top 10% accepted paper)

In this paper, we systematically investigate the inconsistency problems in semi-supervised object detection, where the pseudo boxes may be highly inaccurate and vary greatly at different stages of training. To alleviate the aforementioned problem, we present a holistic semi-supervised object detector termed Consistent-Teacher. Consistent-Teacher achieves compelling improvement on a wide range of evaluations and serves as a new solid baseline for SSOD.

Main Results

All results, logs, configs and checkpoints are listed here. Enjoy 👀!

MS-COCO 1%/2%/5/%/10% Labeled Data

| Method | Data | mAP| config| Links | Google Drive | Baidu Drive |---- | --- |----| ---- | -----| ----|-----| | ConsistentTeacher | MS-COCO 1% | 25.50 | config | log/ckpt |log/ckpt | log/ckpt | ConsistentTeacher | MS-COCO 2% | 30.70 | config | log/ckpt |log/ckpt| log/ckpt | ConsistentTeacher | MS-COCO 5% | 36.60 | config | log/ckpt| log/ckpt| log/ckpt | ConsistentTeacher | MS-COCO 10% | 40.20 | config| log/ckpt| log/ckpt|log/ckpt| | ConsistentTeacher 2x8 | MS-COCO 10% | 38.00 | config| log/ckpt | log/ckpt | log/ckpt | ConsistentTeacher 2x8 (FP16)| MS-COCO 10% | 37.90 | config|log/ckpt |log/ckpt | log/ckpt

MS-COCO100% Labeled + Unlabeled Data

| Method | Data | mAP| config| Links | Google Drive | Baidu Drive |---- | ----| ---- |-----| ----| -----|-----| | ConsistentTeacher 5x8 | MS-COCO 100% + unlabeled |48.20 | config|log/ckpt |log/ckpt| log/ckpt

PASCAL VOC07 Label + VOC12 Unlabel

| Method | Data| mAP| AP50| config| Links |---- | ----| -----| ---- | ---- | ---- | | ConsistentTeacher |PASCAL VOC07 Label + VOC12 Unlabel| 59.00 | 81.00 | config| log/ckpt|

Notes

• Defaultly, all models are trained on 8*V100 GPUs with 5 images per GPU.
• Additionally, we support the 2x8 and fp16 training setting to ensure everyone is able to run the code, even with only 12G graphic cards.
• With 8x2+fp16, the total training time for MS-COCO is less than 1 day.
• We carefully tuned the hyper-parameters after submitting the paper, which is why the results in the repository are slightly higher than those reported in the paper.

Visualizations

Zoom in for better View.

<table> <tbody> <tr> <td><img src="assets/13635702854_d31e5808a5_o_result.png" width="620"></td> <td><img src="assets/8660104283_1012ce0896_z_result.jpg" width="500"></td> </tr> <tr> <td><img src="assets/15678203979_9e85a3f42e_o_results.png" width="620"></td> <td><img src="assets/2_result.png" width="360" float="left"></td> </tr> </tbody> </table>

File Orgnizations

├── configs              
    ├── baseline
    │   |-- mean_teacher_retinanet_r50_fpn_coco_180k_10p.py       
    |       # Mean Teacher COCO 10% config
    |   |-- mean_teacher_retinanet_r50_fpn_voc0712_72k.py      
    |       # Mean Teacher VOC0712 config
    ├── consistent-teacher
    |   |-- consistent_teacher_r50_fpn_coco_360k_fulldata.py           
    |       # Consistent Teacher COCO label+unlabel config
    |
    |   |-- consi