0
$\begingroup$

I trained pre-trained ResNet-101 following caffe model(By Kaiming He github) with 800000 data for training and 200000 data for validation. After I train this model, I got 59% accuracy for 1st and accuracy-top5 is 82% with 30 epoch as seen below picture.

Training process

But when I tried predict some images with this model(net.forward()), the results always produce same probability like below even though I tried with other images.

Produces same output

First thing I thought was image preprocessing problem in predict step like subtracting mean values or adequating batch size corresponding with training step. But all of these step was correctly set up. I checked all other questions having a same problem with me but couldn't find a solution. As the above picture(1) is showing, I assume the training process wasn't something wrong.

I followed "Image Classification and Filter Visualization(00-classification.ipynb)" file provided from Caffe editing model_def and model_weights with my model_def and model_weights having a 30 epoch.

Improve this question
$\endgroup$
3

1 Answer 1

Reset to default
1
$\begingroup$

I solved this problem. The problem was use_global_stats setting in deploy.prototxt. In training step, use_global_stats has to be set as false because mean/var need's to be update. But when I predict using deploy.prototxt use_global_stats has to be set as true.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.