AI Paper Atlas

Research map / Detection and segmentation / Weakly supervised detection and segmentation

SODA10M: A Large-Scale 2D Self/Semi-Supervised Object Detection Dataset for Autonomous Driving

Jianhua Han, Xiwen Liang, Hang Xu, Kai Chen, Lanqing HONG, Jiageng Mao, Chaoqiang Ye, Wei Zhang, Zhenguo Li, Xiaodan Liang, Chunjing Xu

NeurIPS 2021 · Detection and segmentation › Weakly supervised detection and segmentation · approach: Pseudo-labeling for semi-supervised detection · 32 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
pretrain MoCo-v1 on SODA10M instead of BDD100KMoCo-v1 · SODA10M val (OD)mAP: 37.1 -> 38.9Improved
pretrain on BDD100K (smaller, more diverse) instead of nuScenesMoCo-v1 · SODA10M val (OD)mAP: 36.2 -> 37.1Improved
use dense contrastive loss (DenseCL) instead of MoCo-v1 when pretraining on SODA10MFaster RCNN fine-tune · SODA10M valmAP: 38.9 -> 38.1Hurt
use dense contrastive loss (DenseCL) instead of MoCo-v1 when pretraining on ImageNetFaster RCNN fine-tune · SODA10M valmAP: 39.0 -> 39.9Improved
pretrain on SODA10M instead of ImageNetMoCo-v1 · SODA10M val, Faster-RCNNmAP: 39.0 -> 38.9No clear effect
add 5M extra pre-training images (ablation)SimCLR on SODA10M · SODA10M val, Faster-RCNNmAP: 35.9 -> 37.1Improved
add 5M extra pre-training images (ablation)MoCo-v2 on SODA10M · SODA10M val, Faster-RCNNmAP: 38.7 -> 38.6No clear effect
increase pseudo-labeling unlabeled data from 50K to 100K (ablation)Pseudo Labeling · SODA10M valmAP: 39.3 -> 39.9Improved
increase pseudo-labeling unlabeled data from 100K to 500K (ablation)Pseudo Labeling · SODA10M valmAP: 39.9 -> 38.5Hurt
add consistency regularization (STAC) to pseudo labelingFaster RCNN semi-supervised · SODA10M valmAP: 39.9 -> 42.8Improved
use Unbiased Teacher (EMA + focal loss) instead of STACFaster RCNN semi-supervised · SODA10M valmAP: 42.8 -> 46.2Improved
use semi-supervised Unbiased Teacher instead of supervised onlyFaster RCNN · SODA10M night domainmAP: 21.1 -> 39.7Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy