Research map / Vision-language and multimodal / CLIP and zero-shot
Text4Seg++: Advancing Image Segmentation via Generative Language Modeling
Mengcheng Lan, Chaofeng Chen, Jiaxing Xu, Zongrui Li, Yiping Ke, Xudong Jiang, Yingchen Yu, Yunqing Zhao, Song Bai
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 semantic bricks (fg/bg tokens) to box-wise descriptors (ablation) | Text4Seg++ 32x32 B-SD · refCOCO+ (avg over splits) | mIoU: 74.9 -> 75.1 | Improved |
| add semantic bricks (fg/bg tokens) to box-wise descriptors (ablation) | Text4Seg++ 32x32 B-SD · refCOCOg (avg over splits) | mIoU: 78.1 -> 78.3 | Improved |
| add semantic bricks (fg/bg tokens) to box-wise descriptors (ablation) | Text4Seg++ 32x32 B-SD · refCOCO (avg over splits) | mIoU: 79.7 -> 79.7 | No clear effect |
| add semantic bricks to B-SD to cut sequence length (ablation) | Text4Seg++ 64x64 B-SD · RefCOCO token length (Qwen tokenizer) | average tokens: 283.0 -> 150.4 | Improved |
| use box-wise instead of image-wise descriptors at 64x64 (ablation) | Text4Seg++ · RefCOCO token length (Qwen tokenizer) | average tokens: 767.6 -> 283.0 | Improved |
| replace row-wise RLE with image-wise RLE (ablation) | Text4Seg 16x16 I-SD · refCOCO | cIoU: 74.2 -> 70.4 | Hurt |
| replace row-wise RLE with image-wise RLE (ablation) | Text4Seg 16x16 I-SD · refCOCO+ | cIoU: 68.0 -> 64.7 | Hurt |
| add row-wise RLE to full-length semantic descriptors (ablation) | Text4Seg 16x16 I-SD · RefCOCO | average token length: 583 -> 154 | Improved |
| add row-wise RLE to full-length semantic descriptors (ablation) | Text4Seg 16x16 I-SD · RefCOCO | inference time: 19s -> 3x faster | Improved |
| increase B-SD mask resolution from 32x32 up to 80x80 (ablation) | Text4Seg++ · refCOCO/+/g (avg cIoU) | cIoU | Improved |
| increase input image resolution from 560^2 to 896^2 with native aspect ratio (ablation) | Text4Seg++ · refCOCO/+/g (avg cIoU) | cIoU | Improved |
| force the image into a square 896*896 input (ablation) | Text4Seg++ · refCOCO/+/g (avg cIoU) | cIoU | Hurt |
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.
- Text4Seg — Text4Seg: Reimagining Image Segmentation as Text Generation (ICLR 2025) · better on 1 of 1 shared result
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
- Text4Seg: Reimagining Image Segmentation as Text Generation (ICLR 2025 · similarity 0.97)
- Seg4Diff: Unveiling Open-Vocabulary Semantic Segmentation in Text-to-Image Diffusion Transformers (NeurIPS 2025 · similarity 0.90)
- Exploring Simple Open-Vocabulary Semantic Segmentation (CVPR 2025 · similarity 0.88)
- SED: A Simple Encoder-Decoder for Open-Vocabulary Semantic Segmentation (CVPR 2024 · similarity 0.88)
- MaSS13K: A Matting-level Semantic Segmentation Benchmark (CVPR 2025 · similarity 0.88)
- CLIP as RNN: Segment Countless Visual Concepts without Training Endeavor (CVPR 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.
