Open tqvinhcs opened 4 years ago
You can download the pre-trained model through https://drive.google.com/open?id=18QRqGvJygU5ydYxTpEfI6jANR72C_2rc.
I am getting problem in result as mentioned in paper accuracy is 93 but I am getting only 73
Hello, Thank you for your response but after calling above program model run for 2 times with accuracy with model accuracy 41 and 83.85 and average accuracy of 62
Hello, I just download the modeles and test them with the released code, the results are as follows:
Train on 38086 samples, validate on 2005 samples Model Accuracy:88.74 Train on 35763 samples, validate on 1883 samples Model Accuracy:94.28 [88.74, 94.28] ave: 91.50999999999999
According to my experience, the testing results will not change with the released code. Do you change the code?
Thank you for your quick reply. Actually as per the below link of issue I was getting error https://github.com/microsoft/View-Adaptive-Neural-Networks-for-Skeleton-based-Human-Action-Recognition/issues/7#issue-595366294
Due to above error I have made following changes:- In va-cnn.py: I changed worker=0 in place of default=8 in line 43 In data_cnn.py: due to error in scipy.misc imresize function I have used rgb_ske=np.resize(rgb_ske,(224,224,3)) in line number 94.
Could you please guide me how to resolve these issue without changing code as I am using python3.6.
It may caused by the Scipy as mentioned in https://stackoverflow.com/questions/56204985/how-to-fix-scipy-misc-has-no-attribute-imresize/56205147. There is no problem when I use Python 3.6. Therefore, I check the version of Scipy of mine. And the information is as follows: Python 3.6.4 |Anaconda, Inc.| (default, Jan 16 2018, 18:10:19) [GCC 7.2.0] on linux Type "help", "copyright", "credits" or "license" for more information.
import scipy scipy.version '1.0.0'
I have installed scipy 1.1.0 and now it stated running but there I have few more doubts: 1) How can I make this model to run for GPU? 2) In va-rnn I am getting attached error. In va-rnn code I have updated tensor flow as backend in place of theano:
it will be great if it is possible for you to give email address for communication.
For va-cnn, it should run for GPU by default. Could you give me more information? For va-rnn, I tried tensorflow before, but the results are poor and the training speed is slow. Therefore, I suggest to use theano. You can communicate with me through zpengfei@stu.xjtu.edu.cn
can you please provide the pretrained model for VA-RNN also.
can you please provide the pretrained model for VA-RNN also.
You can download both models for VA-RNN and VA-CNN through https://drive.google.com/open?id=12PfMRK2rSg0ndLbTVw-GelRrb2TVxazg
Thakyou for the quick response. Is is possible to view the 3D skeletal poses from view adaption subnetwork as shown in paper?
Hi, I am looking for pre-trained models on NTU CS and CV dataset for the va-cnn. It would be great if you can provide these pre-trained models. Also, do you have code for training on the SYS3DHOI dataset for va-cnn?
Thank you