Open murphypei opened 3 years ago
Our dataset has limited image appearance. Therefore, a model solely trained on our dataset may not generalize well to other in-the-wild images. You can see my recent work (https://arxiv.org/abs/2011.11534) for the in-the-wild hand pose.
Our dataset has limited image appearance. Therefore, a model solely trained on our dataset may not generalize well to other in-the-wild images. You can see my recent work (https://arxiv.org/abs/2011.11534) for the in-the-wild hand pose.
thanks for your reply
I ran the network with default input image, and I got good result as descributions. Then I ran the network with images that come from a video, I find that some frames got good results and other similar frames got bad results. I have checked their bbox, they are similar and correct. It makes me very confused, so can you provide some suggestions?