search

WeiTang114 / MVCNN-TensorFlow

An Multi-View CNN (MVCNN) implementation with TensorFlow.
MIT License
120 stars 67 forks source link

The acc's value is always 50.000000 when training. #25

Open ChengYeung1222 opened 5 years ago

ChengYeung1222 commented 5 years ago

I use two classes(airplane and bathtub) of the modelnet40v1 dataset to train the MVCNN model.I set up 10 shapes per class in the train_lists.txt and 2 shapes per class for validation.Does anyone have the same problem?Is this a random guessing problem?

WeiTang114 commented 5 years ago

Could you try with more training data?

monapr commented 5 years ago

I am having the same problem: I am currently training on 145 shapes and use 45 shapes for validation. While loss decreases, accuracy is always exactly 51.219512 and doesn't change even after 80 epochs. Have you found a solution?

hassaanmahmood commented 5 years ago

I also get this problem but with different dataset. loss is decreasing but accuracy is staying exactly at 50. after training I test the model with test, validation and training datasets, I got 50, 50, 46 respectively. Anyone know what can be the problem? Did anyone check with other implementations?. thanks.