search

agethen / RPAN

Our implementation of Recurrent Pose Attention in Du et al.: "RPAN: An End-to-End Recurrent Pose-attention Network for Action Recognition in Videos"
37 stars 15 forks source link

how to run #1

Closed 13293824182 closed 5 years ago

13293824182 commented 6 years ago

How can I use your code to classify videos?

l2009312042 commented 5 years ago

in data_handle.py ,this line is is error :label[b-start, :] = class_dict[action[0]]

agethen commented 5 years ago

Sorry about that; the uploaded version is not complete, I will upload a fixed version later today. Thank you!

agethen commented 5 years ago

I have uploaded a minimal example run.py, which you need to edit accordingly. You may likely want to add some functionality as well, such as printing results etc (I may add this at a later stage). If there are any further questions, feel free to ask.

l2009312042 commented 5 years ago

thx but i have another question , when load resnet_v2.npy,it fails, and report as follow: Util.py: Loading from resnet_v2.npy NOT loading weights, as file resnet_v2.npy was not found. Please edit util.py. Continuing with random weights. is this file is ok? looking forward to your reply

agethen commented 5 years ago

Hi, Please make sure the path in line 4 in util.py is correct. Try an absolute path if it did not work so-far. If you start out with random weights you will have a hard time (and the ResNet Batch Normalization layers are not setup for learning right now).

l2009312042 commented 5 years ago

thx for your reply .an absolute path is not work , i think the file maybe has some wrong the file video1.npy is ok

agethen commented 5 years ago

Hm, well it's not impossible I suppose, but seems unlikely? Check out if the md5sum is correct, and download the file again if not:

md5sum resnet_v2.npy 
506d0bf5c1cc4fd19b385646153cf2c7  resnet_v2.npy