Closed Xiaolong-RRL closed 1 year ago
Hello Team~ Thanks for your great work!
From the paper, I can see the quantization residual defined as:
but in code, there is little different:
kpt_heatmap_offset_target[batch_id, j, k * 2] = kptx - kptx_int kpt_heatmap_offset_target[batch_id, j, k * 2 + 1] = kpty - kpty_int mask_kpt_heatmap_offset[batch_id, j, k * 2:k * 2 + 2] = 1
which is from: here. It seems like the quantization residual is defined as:
It's different from the formula in the paper, could you tell me if this is true? If true, why to set it this way?
Thank you very much!
Hello Team~ Thanks for your great work!
From the paper, I can see the quantization residual defined as:
but in code, there is little different:
which is from: here. It seems like the quantization residual is defined as:
It's different from the formula in the paper, could you tell me if this is true? If true, why to set it this way?
Thank you very much!