Some problems in data labeling
Opened this issue · 2 comments
SugarMuzi commented
Hi, Thanks for your dataset. I have a question about dataset annotation file.
- I noticed that the frame-level annoations are overlap. for example, A19N03-M.txt, as flows:
10,234,332,2 23,333,506,2 8,517,808,2 39,632,680,2 10,682,737,2 43,739,808,2. the context show us the annotation of 8th action completely masks the annotation of 39th action,10th action and 43th action. so, frames at range 632 to 808 are annoted three action labels. so, if i use the sample to train model, i think model can not recognite
which action is the frame at the range. So, if you are also noticed that?and how to solve this problem?
SugarMuzi commented
Allow me to add that i download PKUMMDv2 datasets
SugarMuzi commented
The number of ovelopping samples are about half of the datasets total