RnDProjectsDeebul / ManojKolpeThesis

Mozilla Public License 2.0
2 stars 1 forks source link

Working code of unet with gp #16

Open Manojkl opened 2 years ago

Manojkl commented 2 years ago

The CUDA overload of memory problem is solved. There was one command missing while validating the model. torch.no_grad(). Because of this, the model's memory is filling up after every iteration and not clearing up. The complete code for training and validation can be found at Link While training the model with 2 sequences and testing on the third sequence following results are obtained. IOU for vanilla unet model: 0.466, u-net-gp model: 0.4452