Xiaomi-Lens-0 adds mask prediction to a 4B Qwen3.5 model
Original title: Xiaomi-Lens-0 Technical Report
A lightweight decoder turns the model's own visual latents into masks, lifting referring comprehension 5.6 points over Qwen3.5.
Xiaomi presents Xiaomi-Lens-0, a 4B multimodal model built on Qwen3.5 that predicts masks from its own visual representations. A lightweight decoder combines vision-encoder patch features with language-model mask-token states, so mask supervision reaches the shared backbone without a separate segmentation encoder. Grounded reasoning traces supervise regions visited during chain-of-thought. Authors report a 5.6-point average gain over Qwen3.5 on eight RefCOCO-family splits and the highest segmentation cIoU among compared models.
Unconfirmed: Benchmark gains are author-reported, and public weight availability is not stated.
- Xiaomi-Lens-0 is a 4B model built on Qwen3.5.
- Referring comprehension rises 5.6 points on average across eight splits.
- Those splits cover RefCOCO, RefCOCO+, and RefCOCOg.
- It reports the highest segmentation cIoU among compared models.
- It also surpasses Qwen3.5 on every perception benchmark the authors cite.
- Vendors
- Xiaomi
- AI models
- Xiaomi-Lens-0Qwen3.5
From the original article
1 min read at huggingface.co
Multimodal large language models (MLLMs) answer questions about whole images well, yet they still struggle with fine-grained region understanding. A key reason is that text-only supervision never specifies which regions support an answer. To close this gap, we present Xiaomi-Lens-0, a 4B MLLM built on Qwen3.
By Pei Fu, Yike Liu, Yue Sun · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.