Research map / Detection and segmentation / Weakly supervised detection and segmentation
Large Scale Visual Recognition through Adaptation using Joint Representation and Multiple Instance Learning
Judy Hoffman, Deepak Pathak, Eric Tzeng, Jonathan Long, Sergio Guadarrama, Trevor Darrell, Kate Saenko
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 |
|---|---|---|---|
| adapt all layers (1-7, fcS) plus background layer with strongly labeled data (ablation) | classification network · ILSVRC13 det val2 (weak categories) | mAP (%): 10.31 -> 15.85 | Improved |
| adapt the background category layer only (ablation) | classification network · ILSVRC13 det val2 (weak categories) | mAP (%): 10.31 -> 12.22 | Improved |
| adapt fc6 in addition to fcbgrnd (ablation) | LSDA rep only · ILSVRC13 det val2 (weak categories) | mAP (%): 12.22 -> 13.72 | Improved |
| adapt fc7 in addition to fcbgrnd (ablation) | LSDA rep only · ILSVRC13 det val2 (weak categories) | mAP (%): 12.22 -> 14.57 | Improved |
| adapt the category-specific layer fcS in addition to fcbgrnd (ablation) | LSDA rep only · ILSVRC13 det val2 (weak categories) | mAP (%): 12.22 -> 11.74 | No clear effect |
| adapt convolutional layers 1-5 in addition to fc layers (ablation) | LSDA rep only · ILSVRC13 det val2 (weak categories) | mAP (%): 14.42 -> 15.85 | Improved |
| add kNN-adapted category-specific layer (k=10) (ablation) | LSDA rep only · ILSVRC13 det val2 (weak categories) | mAP (%): 15.85 -> 16.15 | Improved |
| jointly retrain category-specific weights with discovered boxes (fcW) (ablation) | LSDA rep only · ILSVRC13 det val2 (weak categories) | mAP (%): 15.85 -> 17.01 | Improved |
| jointly retrain all layers and category weights with MIL-discovered boxes (ablation) | LSDA rep only · ILSVRC13 det val2 (weak categories) | mAP (%): 15.85 -> 18.08 | Improved |
| add strongly labeled auxiliary data adaptation | classification network · ILSVRC13 det val1 discovered boxes | localization mAP (%) ov=0.5: 13.13 -> 22.81 | Improved |
| add strongly labeled auxiliary data adaptation | classification network · ILSVRC13 det val1 discovered boxes | CorLoc (%) ov=0.5: 26.10 -> 28.81 | Improved |
| adapt the network with LSDA instead of classification-only network for segmentation | FCN-32s AlexNet · PASCAL VOC 2011 val segmentation | mean IU (%): 11.6 -> 21.3 | Improved |
Closest papers by meaning
- Train a One-Million-Way Instance Classifier for Unsupervised Visual Representation Learning (AAAI 2021 · similarity 0.90)
- [CLS] is Not Enough: Multi-Label Recognition via Patch-Level Inference and Adaptive Aggregation (ICML 2026 · similarity 0.89)
- Contextual Convolutional Networks (ICLR 2023 · similarity 0.89)
- Exploit Bounding Box Annotations for Multi-Label Object Recognition (CVPR 2016 · similarity 0.89)
- Is Second-Order Information Helpful for Large-Scale Visual Recognition? (ICCV 2017 · similarity 0.89)
- A Multi-Modal, Discriminative and Spatially Invariant CNN for RGB-D Object Labeling (TPAMI 2017 · similarity 0.89)
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.
