search

shekkizh / FCN.tensorflow

Tensorflow implementation of Fully Convolutional Networks for Semantic Segmentation (http://fcn.berkeleyvision.org)
MIT License
1.25k stars 527 forks source link

i made a reimplement with tensorflow2 #127

Open YuhangJi opened 2 years ago

YuhangJi commented 2 years ago

i made a reimplement with tensorflow2, but it is not very well. In MIT dataset, the model loss value just decline to 2.3.

varungupta31 commented 2 years ago

Hey @YuhangJi, I'm tring to train the FCN on my custom dataset, in which I have only 2 classes (background + object). Can you guide me with how to manipulate the code to achieve this?

TIA

YuhangJi commented 2 years ago

i made a reimplement with tensorflow2, but it is not very well. In MIT dataset, the model loss value just decline to 2.3.

This issue is caused by the absence of a softmax function when training a model.

YuhangJi commented 2 years ago

Hey @YuhangJi, I'm tring to train the FCN on my custom dataset, in which I have only 2 classes (background + object). Can you guide me with how to manipulate the code to achieve this?

TIA

No problem. @varungupta31