Research map / Vision-language and multimodal / CLIP and zero-shot
SLAN: Self-Locator Aided Network for Vision-Language Understanding
Jiang-Tian Zhai, Qi Zhang, Tong Wu, Xing-Yu Chen, Jiang-Jiang Liu, Ming-Ming Cheng
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 frozen COCO-pretrained region proposal module with learnable self-locator (ablation) | SLAN ViT-Base · COCO and Flickr30k retrieval | TR@1 / IR@1: 68.5 -> 69.1 (COCO TR@1); 74.1 -> 76.2 (Flickr30k IR@1) | Improved |
| increase number of region adaptors from 0 to 3 (ablation) | SLAN ViT-Base · COCO retrieval | TR@1: 69.1 -> 72.1 | Improved |
| increase number of region adaptors from 0 to 3 (ablation) | SLAN ViT-Base · Flickr30k retrieval | IR@1: 76.2 -> 78.9 | Improved |
| keep only top-K regions ranked by saliency (ablation) | SLAN region filter · COCO retrieval | TR@1: 69.4 -> 70.6 | Improved |
| add saliency score threshold to top-K region filter (ablation) | SLAN region filter · COCO retrieval | TR@1: 70.6 -> 72.1 | Improved |
| use SLAN with ViT-Huge versus Beit-3 on VQAv2 | SLAN · VQAv2 test-dev | VQA score: 84.2 -> 84.5 | Improved |
| use SLAN versus Beit-3 on NLVR2 | SLAN · NLVR2 dev | accuracy: 91.5 -> 91.0 | No clear effect |
| scale vision backbone from ResNet50 to ViT-Huge (ablation) | SLAN · COCO object detection fine-tune | AP: 59.2 -> 63.5 | Improved |
| use SLAN with ResNet50 versus larger vision backbones for retrieval | SLAN ResNet50 · COCO retrieval | TR@1: ResNet50 85.1 vs BLIP ViT-Large 82.4 with fewer parameters | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Beit-3 — BEiT: BERT Pre-Training of Image Transformers (ICLR 2022) · better on 7 of 10 shared results
- BLIP — BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generation (ICML 2022) · better on 4 of 4 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
- Grounding Everything: Emerging Localization Properties in Vision-Language Transformers (CVPR 2024 · similarity 0.89)
- RegionGPT: Towards Region Understanding Vision Language Model (CVPR 2024 · similarity 0.88)
- SC-Tune: Unleashing Self-Consistent Referential Comprehension in Large Vision Language Models (CVPR 2024 · similarity 0.88)
- Locality Alignment Improves Vision-Language Models (ICLR 2025 · similarity 0.88)
- Recognize Any Regions (NeurIPS 2024 · similarity 0.87)
- Enhancing Large Vision Language Models with Self-Training on Image Comprehension (NeurIPS 2024 · similarity 0.87)
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.
