Research map / Adversarial robustness and security / Attacks on text and graphs
Robust onion: Peeling Open Vocab Object Detectors Under Noise
Priyank Pathak, Mukilan Karuppasamy, Aaditya Baranwal, Shruti Vyas, Yogesh Rawat
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 trainable spatial tokens (TK0) to frozen GLIP-T backbone | GLIP-T · BDD-100K trained, All | mAP/AP50 (%): 31.4 -> 45.1 | Improved |
| add TK0 tokens to frozen model | GLIP-T · COCO zero-shot | mAP (%): 46.6 -> 43.2 | Hurt |
| end-to-end finetuning instead of frozen model | GLIP-T · COCO zero-shot | mAP (%): 46.6 -> 3.0 | Hurt |
| train only fusion transformer (Fuse) instead of full E2E | GLIP-T · BDD-100K trained, All | mAP/AP50 (%): 66.4 -> 65.1 | No clear effect |
| combine NN cross-exchange block with TK0 (ablation) | GLIP-T · BDD-100K trained, All | mAP/AP50 (%): 45.1 -> 55.2 | Improved |
| combine NN and TK0 versus TK0 alone (ablation) | GLIP-T · COCO zero-shot | mAP (%): 43.2 -> 33.4 | Hurt |
| NN & TK0 instead of zero-shot | GLIP-T · Foggy Cityscapes zero-shot | mAP (%): 21.4 -> 28.2 | Improved |
| NN & TK0 instead of zero-shot | GLIP-T · Virtual KITTI 2 zero-shot (All) | mAP (%): 20.9 -> 29.9 | Improved |
| NN & TK0 instead of zero-shot | GLIP-T · Flickr30k -> WiderFace AP | AP: 11.98 -> 13.23 | Improved |
| NN & TK0 instead of zero-shot | GLIP-T · UAVDT -> VisDRONE AP | AP: 20.44 -> 19.34 | Hurt |
| NN & TK0 (2.41M params) instead of E2E (231.76M) | GLIP-T · trainable parameters | params (M): 231.76 -> 2.41 | Improved |
| add LLM-modified noise-aware captions at evaluation | GLIP variants · Flickr30k pixelation | Recall@1 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Fuse — FUSE: Full‑spectrum Unlearnable Examples via Spectral Equalization (ICML 2026) · better on 4 of 5 shared results
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
- OW-OVD: Unified Open World and Open Vocabulary Object Detection (CVPR 2025 · similarity 0.87)
- How to Evaluate the Generalization of Detection? A Benchmark for Comprehensive Open-Vocabulary Detection (AAAI 2024 · similarity 0.86)
- Language-conditioned Detection Transformer (CVPR 2024 · similarity 0.86)
- OVT-B: A New Large-Scale Benchmark for Open-Vocabulary Multi-Object Tracking (NeurIPS 2024 · similarity 0.85)
- VLM-RobustBench: A Comprehensive Benchmark for Robustness of Vision-Language Models (ICML 2026 · similarity 0.85)
- COUNTS: Benchmarking Object Detectors and Multimodal Large Language Models under Distribution Shifts (CVPR 2025 · 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.
