[NeurIPS 2021 Spotlight] Aligning Pretraining for Detection via Object-Level Contrastive Learning

By Fangyun Wei*, Yue Gao*, Zhirong Wu, Han Hu, Stephen Lin.

* Equal contribution.


Image-level contrastive representation learning has proven to be highly effective as a generic model for transfer learning. Such generality for transfer learning, however, sacrifices specificity if we are interested in a certain downstream task. We argue that this could be sub-optimal and thus advocate a design principle which encourages alignment between the self-supervised pretext task and the downstream task. In this paper, we follow this principle with a pretraining method specifically designed for the task of object detection. We attain alignment in the following three aspects:

  1. object-level representations are introduced via selective search bounding boxes as object proposals;
  2. the pretraining network architecture incorporates the same dedicated modules used in the detection pipeline (e.g. FPN);
  3. the pretraining is equipped with object detection properties such as object-level translation invariance and scale invariance. Our method, called Selective Object COntrastive learning (SoCo), achieves state-of-the-art results for transfer performance on COCO detection using a Mask R-CNN framework.


Main results

The pretrained models will be available soon.

SoCo pre-trained models

Model Arch Epochs Scripts Download
SoCo ResNet50-C4 100 SoCo_C4_100ep
SoCo ResNet50-C4 400 SoCo_C4_400ep
SoCo ResNet50-FPN 100 SoCo_FPN_100ep
SoCo ResNet50-FPN 400 SoCo_FPN_400ep
SoCo* ResNet50-FPN 400 SoCo_FPN_Star_400ep

Results on COCO with MaskRCNN R50-FPN

Methods Epoch APbb APbb50 APbb75 APmk APmk50 APmk75 Detectron2 trained
Scratch 31.0 49.5 33.2 28.5 46.8 30.4
Supervised 90 38.9 59.6 42.7 35.4 56.5 38.1
SoCo 100 42.3 62.5 46.5 37.6 59.1 40.5
SoCo 400 43.0 63.3 47.1 38.2 60.2 41.0
SoCo* 400 43.2 63.5 47.4 38.4 60.2 41.4

Results on COCO with MaskRCNN R50-C4

Methods Epoch APbb APbb50 APbb75 APmk APmk50 APmk75 Detectron2 trained
Scratch 26.4 44.0 27.8 29.3 46.9 30.8
Supervised 90 38.2 58.2 41.2 33.3 54.7 35.2
SoCo 100 40.4 60.4 43.7 34.9 56.8 37.0
SoCo 400 40.9 60.9 44.3 35.3 57.5 37.3

Get started


The Dockerfile is included, please refer to it.

Prepare data with Selective Search

  1. Generate Selective Search proposals

    python selective_search/
  2. Filter out not valid proposals with filter strategy

    python selective_search/
  3. Post preprocessing for no proposals images

    python selective_search/

Pretrain with SoCo

Use SoCo FPN 100 epoch as example.

bash ./tools/

Finetune detector

  1. Copy the folder detectron2_configs to the root folder of Detectron2
  2. Train the detectors with Detectron2


  title={Aligning Pretraining for Detection via Object-Level Contrastive Learning},
  author={Wei, Fangyun and Gao, Yue and Wu, Zhirong and Hu, Han and Lin, Stephen},
  journal={arXiv preprint arXiv:2106.02637},


View Github