Closed CHENYUHANG-smp closed 2 years ago
Hi, The focal length of 5000 is indeed not accurate. There are papers trying to tackle this problem:
Beyond Weak Perspective for Monocular 3D Human Pose Estimation, ECCV 2020 SPEC: Seeing People in the Wild with an Estimated Camera, ICCV 2021
Thanks for your brilliant work! I notice that you estimate the camera parameters by using a function called "estimate_translation_np" (borrowed from SPIN). I feel puzzled that why you could set the same focal length (5000) for different datasets and different images. Thanks again and hope for your reply.