search

qqwweee / keras-yolo3

A Keras implementation of YOLOv3 (Tensorflow backend)
MIT License
7.14k stars 3.44k forks source link

training tiny-yolov3 problems #280

Open mechealww2 opened 5 years ago

mechealww2 commented 5 years ago

Create Tiny YOLOv3 model with 6 anchors and 8 classes. C:\Users\LYK_pc\Anaconda3\lib\site-packages\keras\engine\saving.py:1140: UserWarning: Skipping loading of weights for layer conv2d_10 due to mismatch in shape ((1, 1, 512, 39) vs (255, 512, 1, 1)). weight_values[i].shape)) C:\Users\LYK_pc\Anaconda3\lib\site-packages\keras\engine\saving.py:1140: UserWarning: Skipping loading of weights for layer conv2d_10 due to mismatch in shape ((39,) vs (255,)). weight_values[i].shape)) C:\Users\LYK_pc\Anaconda3\lib\site-packages\keras\engine\saving.py:1140: UserWarning: Skipping loading of weights for layer conv2d_13 due to mismatch in shape ((1, 1, 256, 39) vs (255, 256, 1, 1)). weight_values[i].shape)) C:\Users\LYK_pc\Anaconda3\lib\site-packages\keras\engine\saving.py:1140: UserWarning: Skipping loading of weights for layer conv2d_13 due to mismatch in shape ((39,) vs (255,)). weight_values[i].shape))

Why there are mismatch?And I do not get good performance on my own dataset.Training stopped at val_loss =27... Anyone can help? Thks.

samuelefiorini commented 5 years ago

Hi, I am experiencing the same issue. Any news on that?

mechealww2 commented 5 years ago

Sorry，I do not find any solution. But i find keras yolov2 and ssd have better performance on my task.

samuelefiorini commented 5 years ago

Thanks @mechealww2, which implementation of yolov2 are you using?

mechealww2 commented 5 years ago

https://github.com/experiencor/keras-yolo2

DaliborCimr commented 2 years ago

The mismatch is because you are trying to train your own dataset with 8 classes. Last layer output depend on it because of equation "num_anchors*(num_classes+5)" which is in darknet model 3x(80+5)=255 and in your case 3x(8+5)=39