Research map / Label-efficient and robust learning / Self-supervised and contrastive
Exploiting Unlabeled Data in CNNs by Self-Supervised Learning to Rank
Xialei Liu, Joost van de Weijer, Andrew D. Bagdanov
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 |
|---|---|---|---|
| add multi-task ranking loss with efficient Siamese backprop (vs RankIQA+FT) (ablation) | MT-RankIQA · TID2013 (entire, 80/20 split) | LCC / SROCC: further improving the accuracy by 2.8% on LCC and 2.6% on SROCC | Improved |
| add ranking pretraining then fine-tune (vs direct fine-tuning) (ablation) | RankIQA+FT · TID2013 (entire, 80/20 split) | LCC: we improve on the baseline by 11% | Improved |
| use efficient Siamese sampling instead of random pairs (ablation) | RankIQA+FT · TID2013 (entire, 80/20 split) | LCC: 0.775 -> 0.799 | Improved |
| use efficient Siamese sampling instead of hard negative mining (ablation) | MT-RankIQA · TID2013 (entire, 80/20 split) | LCC: 0.810 -> 0.827 | Improved |
| add more synthetic distortions as auxiliary data (ablation) | MT-RankIQA · TID2013 unseen distortions | average gain (unseen): +0.016 -> +0.075 | Improved |
| add ImageNet pre-training (ablation) | Basic CNN · UCF_CC_50 | Ave MAE: 454.45 -> 328.54 | Improved |
| add multi-scale augmentation on pretrained CNN (ablation) | Basic CNN · UCF_CC_50 | Ave MAE: 328.54 -> 308.06 | Improved |
| ranking pretrain + fine-tune instead of direct fine-tuning from ImageNet (ablation) | Ranking+FT · UCF_CC_50 | Ave MAE: 308.06 -> 325.73 | Hurt |
| add multi-task ranking+counting with random pairs (ablation) | Multi-task · UCF_CC_50 | Ave MAE: 325.73 -> 292.67 | Improved |
| use efficient backprop instead of random pair sampling (ablation) | Multi-task · UCF_CC_50 | Ave MAE: 292.67 -> 279.60 | Improved |
| use Query-by-example instead of Keyword unlabeled data (ablation) | Multi-task · UCF_CC_50 | MAE: 279.6 -> 291.5 | Hurt |
| use Query-by-example instead of Keyword unlabeled data (ablation) | Multi-task · ShanghaiTech | MAE: 73.6 -> 72.0 | Improved |
Closest papers by meaning
- Big Self-Supervised Models are Strong Semi-Supervised Learners (NeurIPS 2020 · similarity 0.87)
- RankIQA: Learning From Rankings for No-Reference Image Quality Assessment (ICCV 2017 · similarity 0.87)
- Leveraging Unlabeled Data for Crowd Counting by Learning to Rank (CVPR 2018 · similarity 0.86)
- Automatically Discovering and Learning New Visual Categories with Ranking Statistics (ICLR 2020 · similarity 0.85)
- Universum Prescription: Regularization Using Unlabeled Data (AAAI 2017 · similarity 0.85)
- Small Data Challenges in Big Data Era: A Survey of Recent Progress on Unsupervised and Semi-Supervised Methods (TPAMI 2020 · similarity 0.85)
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.
