SuperAPI AI Directory

AI developer platforms

3,641 results · Page 101 of 102

Visual-RFT

AI developer platforms

Visual-RFT is the official code repository for Visual Reinforcement Fine-Tuning, and the first work to fully adapt DeepSeek-R1's reinforcement learning strategy to the multimodal domain. Using Qwen2-VL-2/7B as the base model, it designs rule-based verifiable rewards and integrates them into a GRPO-based reinforcement fine-tuning framework to improve the performance of large vision-language models on a variety of visual perception tasks. This work has been accepted by ICCV 2025, and the training code, training data, and evaluation scripts have been open-sourced. The follow-up work Visual-ARFT further endows open-source LVLMs with agentic capabilities such as browsing websites and writing code to process images, and proposes the multimodal agent tool benchmark MAT.

Read more