Research map / Label-efficient and robust learning / Noisy labels
MoPro: Webly Supervised Learning with Momentum Prototypes
Junnan Li, Caiming Xiong, Steven Hoi
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 changed | Where | Measured | Effect |
|---|---|---|---|
| Use MoPro instead of Cross-Entropy pretraining | ResNet-50 pretrained on WebVision-V1.0 · WebVision-V1.0 clean validation | top-1 accuracy: 72.4 -> 73.9 | Improved |
| Use MoPro instead of Cross-Entropy pretraining | ResNet-50 pretrained on WebVision-V1.0 · ILSVRC 2012 validation | top-1 accuracy: 65.7 -> 67.8 | Improved |
| Use MoPro instead of Cross-Entropy pretraining | ResNet-50 pretrained on WebVision-V1.0 · VOC07, 1-shot | mAP: 54.5 -> 59.5 | Improved |
| Use MoPro instead of Cross-Entropy pretraining | ResNet-50 pretrained on WebVision-V1.0 · ImageNet, 1% finetuning | top-1 accuracy: 67.6 -> 71.2 | Improved |
| Use MoPro instead of Cross-Entropy pretraining | ResNet-50 pretrained on WebVision-V1.0 · COCO val2017, 1× schedule | bounding-box AP: 39.2 -> 39.7 | Improved |
| Use MoPro instead of Cross-Entropy pretraining | ResNet-50 pretrained on WebVision-V1.0 · ImageNet-R | accuracy: 49.56 -> 54.87 | Improved |
| Add prototypical contrastive loss (ablation) | MoPro pretrained on WebVision-V0.5 · ImageNet, 1% finetuning | top-1 accuracy: 68.0 -> 69.3 | Improved |
| Add instance contrastive loss (ablation) | MoPro pretrained on WebVision-V0.5 · ImageNet, 1% finetuning | top-1 accuracy: 68.2 -> 69.3 | Improved |
| Add both prototypical and instance contrastive losses (ablation) | MoPro pretrained on WebVision-V0.5 · ImageNet, 1% finetuning | top-1 accuracy: 66.9 -> 69.3 | Improved |
| Use prototype similarity in pseudo-labeling (ablation) | MoPro pretrained on WebVision-V0.5 · ImageNet, 1% finetuning | top-1 accuracy: 68.4 -> 69.3 | Improved |
Later papers that compared with this method
4 later accepted papers report MoPro† as a baseline in their results tables; 4 of them report a better number on most of the results they share with it.
- FoPro: Few-Shot Guided Robust Webly-Supervised Prototypical Learning (AAAI 2023 · better on 3 of 3 shared results)
- CAPro: Webly Supervised Learning with Cross-modality Aligned Prototypes (NeurIPS 2023 · better on 1 of 1 shared result)
- GenKL: An Iterative Framework for Resolving Label Ambiguity and Label Non-conformity in Web Images Via a New Generalized KL Divergence (IJCV 2023 · better on 1 of 1 shared result)
- Selective-Supervised Contrastive Learning With Noisy Labels (CVPR 2022 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Intermediate Layers Matter in Momentum Contrastive Self Supervised Learning (NeurIPS 2021 · similarity 0.85)
- FoPro: Few-Shot Guided Robust Webly-Supervised Prototypical Learning (AAAI 2023 · similarity 0.85)
- Momentum Contrast for Unsupervised Visual Representation Learning (CVPR 2020 · similarity 0.85)
- Fast-MoCo: Boost Momentum-Based Contrastive Learning with Combinatorial Patches (ECCV 2022 · similarity 0.84)
- Webly Supervised Fine-Grained Recognition: Benchmark Datasets and an Approach (ICCV 2021 · similarity 0.83)
- Scaling and Benchmarking Self-Supervised Visual Representation Learning (ICCV 2019 · similarity 0.83)
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.
