Research map / Vision-language and multimodal / CLIP and zero-shot
VLOD-TTA: Test-Time Adaptation of Vision-Language Object Detectors
Atif Belal, Heitor Medeiros, Marco Pedersoli, Eric Granger
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 |
|---|---|---|---|
| replace standard entropy with IoU-weighted entropy (IWE) (ablation) | Adapter (YOLO-World) · Watercolor, ClipArt, Comic | mAP / AP50 | Improved |
| add image-conditioned prompt selection (IPS) on top of IWE (ablation) | VLOD-TTA (YOLO-World) · Watercolor, ClipArt, Comic | mAP / AP50 | Improved |
| use prompt averaging instead of ZS prompts (ablation) | YOLO-World · Watercolor and Comic | AP50: -0.35 on Watercolor, -0.30 on Comic | Hurt |
| use prompt selection instead of averaging (ablation) | YOLO-World · Watercolor, ClipArt, Comic | AP50: average gain of +1.0 over ZS | Improved |
| use CLIP-style prompts instead of GPT prompts (ablation) | VLOD-TTA (YOLO-World) · ClipArt, Comic | AP50: -0.8 on ClipArt, -1.3 on Comic | Hurt |
| use dataset-specific prompts instead of dataset-agnostic GPT prompts (ablation) | VLOD-TTA (YOLO-World) · Watercolor, ClipArt, Comic | AP50: +0.9 -> +1.0 Watercolor, +1.0 -> +1.6 ClipArt, +1.2 -> +1.7 Comic | Improved |
| adapt vision backbone with adapters (ablation) | YOLO-World · style-shift datasets avg | AP50: +4.4 AP50 | Improved |
| adapt text encoder with adapters (ablation) | Grounding DINO · style-shift datasets avg | AP50: +3.3 AP50 | Improved |
| adapt vision encoder or neck with adapters (ablation) | Grounding DINO · style-shift datasets avg | AP50 | Hurt |
| add single scale augmentation (ablation) | VLOD-TTA (YOLO-World) · Watercolor, ClipArt, Comic | AP50: +6.0, +6.6, +5.1 | Improved |
| add more than one augmentation (ablation) | VLOD-TTA (YOLO-World) · Watercolor, ClipArt, Comic | AP50 | No clear effect |
| increase input resolution from 640x640 to 1024x1024 (ablation) | VLOD-TTA (YOLO-World) · Cityscapes | mAP gain over ZS: +0.6 mAP -> +2.7 mAP | Improved |
Closest papers by meaning
- The Illusion of Progress? A Critical Look at Test-Time Adaptation for Vision-Language Models (NeurIPS 2025 · similarity 0.93)
- Frustratingly Easy Test-Time Adaptation of Vision-Language Models (NeurIPS 2024 · similarity 0.91)
- Efficient Test-time Adaptation through Candidate Verification and Divergence Shifts (NeurIPS 2026 · similarity 0.90)
- Test-Time Adaptation of Vision-Language Models for Open-Vocabulary Semantic Segmentation (NeurIPS 2025 · similarity 0.90)
- Noisy Test-Time Adaptation in Vision-Language Models (ICLR 2025 · similarity 0.90)
- Roboflow100-VL: A Multi-Domain Object Detection Benchmark for Vision-Language Models (NeurIPS 2025 · similarity 0.90)
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.
