FloodVL-Pix
FloodVL-Pix: Language-Guided Pixel-Level Flood Understanding Benchmark
A large-scale remote sensing benchmark with 131,200 image patches and 443,100 textual descriptions for language-level flood understanding and pixel-level segmentation.
How to run
$ npm i -g lettokenburn-cli$ ltb bench info floodvl-pix
External harness. No Harbor adapter has been generated for this benchmark yet. `ltb bench info` prints the known links.
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
Scores
11 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| Kimi 2.6 | 0.6 | N/A | Independent | medium | Oct 16, 2026 | arxiv.org |
| Qwen3-VL-32B | 0.5 | N/A | Independent | medium | Oct 16, 2026 | arxiv.org |
| GLM-5.3 Flash | 0.5 | N/A | Independent | medium | Oct 16, 2026 | arxiv.org |
| Doubao Seed 2.0 Pro | 0.4 | N/A | Independent | medium | Oct 16, 2026 | arxiv.org |
| DeepSeek V4 Flash | 0.4 | N/A | Independent | medium | Oct 16, 2026 | arxiv.org |
| Doubao-Seed-2.0-Lite | 0.4 | N/A | Independent | medium | Oct 16, 2026 | arxiv.org |
| Doubao Seed 2.1 Turbo | 0.4 | N/A | Independent | medium | Oct 16, 2026 | arxiv.org |
| Qwen3-VL-2B | 0.4 | N/A | Independent | medium | Oct 16, 2026 | arxiv.org |
| Qwen3-VL-4B | 0.3 | N/A | Independent | medium | Oct 16, 2026 | arxiv.org |
| TeoChat-7B | 0.2 | N/A | Independent | medium | Oct 16, 2026 | arxiv.org |
| GeoPixel-7B | 0.1 | N/A | Independent | medium | Oct 16, 2026 | arxiv.org |
Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.
Reproduced by the community
All runsAbout
FloodVL-Pix comprises 106,500 SAR and 24,700 optical image patches with multi-temporal annotations. It supports both language-level severity/evolution descriptions and pixel-level flood segmentation tasks.
The benchmark addresses attentional bias in VLMs by including abundant large-flood samples. Experiments show that the proposed FloodVLM-Pix model improves description quality and segmentation performance compared to state-of-the-art models.