YOLOER V2 tren Hugging Face Spaces (CPU realtime webcam). Object detection da chuyen sang YOLOE (THU-MIG) de tang do chinh xac. Engine su dung adaptive infer (fast frame + hires refresh) va depth async de giam lag. Co the nhap prompt classes de tap trung vao nhom vat the muc tieu. Co the bat MiDaS v2.1 Small de phan tich depth va toa do 3D theo thoi gian thuc.