Dimensions overall score 8.0

GitHub Code Pulse

No public code linked for this paper yet.

• Ostrakon-VL achieves an average score of 60.1 on the ShopBench benchmark, establishing a new state of the art among open-source MLLMs with comparable parameter scales.result 95%
• Ostrakon-VL surpasses the substantially larger Qwen3-VL-235B-A22B model (which scored 59.4) by +0.7 points on ShopBench.result 95%
• Ostrakon-VL exceeds the same-scale Qwen3-VL-8B model (which scored 55.3) by +4.8 points on ShopBench, demonstrating significantly improved parameter efficiency.result 95%
• ShopBench is the first public benchmark specifically designed for Food-Service and Retail Stores (FSRS) scenarios.technical 90%
• The QUAD (Quality-aware Unbiased Automated Data-curation) pipeline is a multi-stage multimodal instruction data curation method developed to address noisy FSRS data.method 85%
• Real-world FSRS data collected from heterogeneous acquisition devices are highly noisy and lack auditable, closed-loop data curation, impeding construction of high-quality training corpora.limitation 90%
• Existing evaluation protocols lack a unified, fine-grained and standardized benchmark spanning single-image, multi-image, and video inputs for FSRS scenarios.limitation 85%
• Ostrakon-VL is based on the Qwen3-VL-8B architecture and specifically designed for Food-Service and Retail Stores applications.technical 90%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card