Skyra 复现：基于 Qwen2.5-VL 的 AI 生成视频检测，SFT(LoRA) + GRPO 两阶段训练，含分布式配置、评测和端到端推理 | Reproduction of Skyra (CVPR 2026) with SFT+GRPO training
reinforcement-learning lora skyra deepfake-detection chain-of-thought supervised-finetuning vllm fsdp multimodal-llm llama-factory ai-generated-video aigc-detection qwen2-vl grpo verl video-forensics cvpr2026 artifact-reasoning
-
Updated
Aug 26, 2026 - Python