Research map / Detection and segmentation / Open-set and few-shot detection
What Makes Instance Discrimination Good for Transfer Learning?
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Filter negatives using labels without pulling same-class instances together | MoCo-v2 (Exemplar-v2) · VOC07 | AP: 48.5 -> 48.8 | Helped |
| Filter negatives using labels without pulling same-class instances together | MoCo-v2 (Exemplar-v2) · ImageNet | Acc: 67.5 -> 68.9 | Helped |
| Filter negatives using labels without pulling same-class instances together | MoCo-v2 (Exemplar-v2) · COCO | AP_box: 38.7 -> 39.4 | Helped |
| Filter negatives using labels without pulling same-class instances together | MoCo-v2 (Exemplar-v2) · COCO | AP_seg: 34.0 -> 34.4 | Helped |
| Filter negatives using labels without pulling same-class instances together | MoCo-v2 (Exemplar-v2) · Cityscapes | mIoU: 78.6 -> 78.8 | Helped |
| Add color jittering during supervised pretraining (ablation) | Supervised pretraining · VOC07 | AP: 43.4 -> 45.9 | Helped |
| Add Gaussian blur during supervised pretraining (ablation) | Supervised pretraining · COCO | AP_box: 39.3 -> 38.9 | Hurt |
| Add Gaussian blur during unsupervised pretraining (ablation) | Unsupervised pretraining · VOC07 | AP: 47.7 -> 48.5 | Helped |
| Pretrain on Synthia rather than ImageNet (ablation) | MoCo-v2 · VOC07 | AP: 48.5 -> 37.4 | Hurt |
| Increase supervised pretraining from 90 to 800 epochs (ablation) | Supervised pretraining · VOC07 | AP: 45.4 -> 47.6 | Helped |
| Use τ=0.1 for Exemplar-v2 (ablation) | Exemplar-v2 · ImageNet | Acc: 68.1 -> 68.9 | Helped |
| Use MoCo-v2 instead of supervised pretraining | Mini-ImageNet few-shot transfer · Mini-ImageNet base classes | Acc: 83.6 -> 75.3 | Hurt |
Closest papers by meaning
- Boost Supervised Pretraining for Visual Transfer Learning: Implications of Self-Supervised Contrastive Representation Learning (AAAI 2022 · similarity 0.90)
- Revisiting the Transferability of Supervised Pretraining: An MLP Perspective (CVPR 2022 · similarity 0.90)
- How Well Do Self-Supervised Models Transfer? (CVPR 2021 · similarity 0.88)
- Aligning Pretraining for Detection via Object-Level Contrastive Learning (NeurIPS 2021 · similarity 0.88)
- On the Surprising Effectiveness of Attention Transfer for Vision Transformers (NeurIPS 2024 · similarity 0.87)
- Understanding the Transferability of Representations via Task-Relatedness (NeurIPS 2024 · similarity 0.86)
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.