AI Paper Atlas

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

Contrastive Deep Supervision

Linfeng Zhang, Xin Chen, Junbo Zhang, Runpei Dong, Kaisheng Ma

ECCV 2022 · Detection and segmentation › Weakly supervised detection and segmentation · approach: Weakly supervised detection · 57 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
add contrastive losses on intermediate layersResNet18 · CIFAR100top-1 accuracy (%): 77.45 -> 80.84Improved
add contrastive losses on intermediate layersResNet18 · ImageNettop-1 accuracy (%): 69.21 -> 72.85Improved
pretrain backbone with contrastive deep supervision before detectionFaster RCNN ResNet50 · COCO2017AP: 37.4 -> 38.3Improved
pretrain backbone with contrastive deep supervision before detectionRetinaNet ResNet50 · COCO2017AP: 36.5 -> 37.3Improved
pretrain backbone with cross-entropy deep supervision (DSN) instead of standard trainingFaster RCNN ResNet50 · COCO2017AP: 37.4 -> 37.3No clear effect
add contrastive deep supervision, training from scratchResNet50 · CUB200 (from scratch)top-1 accuracy (%): 60.65 -> 64.65Improved
add contrastive deep supervision, finetuning from ImageNetResNet50 · CUB200 (finetuned)top-1 accuracy (%): 78.50 -> 82.10Improved
add contrastive deep supervision to feature distillationResNet18 student · ImageNettop-1 accuracy (%): 69.21 -> 73.23Improved
distill projection-head embeddings rather than backbone features (ablation)CDS distillation · CIFAR100top-1 accuracy (%): 82.26 -> 83.53Improved
vary where projection heads are placed (uniform, downsampling, shallow, deep) (ablation)CDS · CIFAR100 ResNet50top-1 accuracy (%): 81.23, 81.31, 81.07, 80.99No clear effect
add more projection heads up to five (ablation)CDS · CIFAR100top-1 accuracy (%)Mixed
use contrastive deep supervision with AutoAugment and 350 epochs versus SupConResNet50 · ImageNettop-1 accuracy (%): 78.7 -> 79.8Improved

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