SemTorch
SemTorch copied to clipboard
WaterKnight1998
Hi Thank you for the great repo. After training the model i try to save it by: aux=learn.model aux=aux.cpu() traced_cell=torch.jit.script(aux) traced_cell.save("whatever.pth") learn.save('whatever') But unfortunately the kernel dies each time! Is...
Hello, I am trying to use SemTorch for instance segmentation for the case when multiple masks are present in the same image. By looking at the source code and the[...
Multiple points in the application have been changed to match hrnet_w18_small_v1 with hrnet_w18_small_model_v1 and with v2 the same. Now the training works fine with these small models.
With `get_segmentation_learner(architecture_name='hrnet', backbone_name='hrnet_w18')`. Using the following callback to save the models during training: `SaveModelCallback(monitor='dice_multi', fname='best_model', with_opt=True)` The results of the predictions after loading "best_model.pth" with `learner.load` are zero-filled masks. The...
Hi, when using get_segmentation_learner function with architecture_name="unet" and backbone_name="resnet50" I'm getting "RuntimeError: cuDNN error: CUDNN_STATUS_NOT_INITIALIZED" error. I'm using torch 1.7.1, torchvision 0.8.2 and cuda 10.2 because of fastai compatibility. Do...
In the fastai v2 `unet_learner` function call, n_in can be specified in order to reshape the learner to accept inputs that don't have 3 channels. Is this something that can...
Since it seems that `Mish` activation works nicely in Segmentation tasks. I would like to add it to the learner as performed [here](https://muellerzr.github.io/fastblog/papers/2020/09/18/CAMVID.html#The-Model-and-Training). I will take a look how this...
Hi, The numpy aliases for builtin types (e.g. np.int) have been deprecated. This breaks the instantiation of the HRNet model. The deprecation was announced in [Numpy 1.20 Release Notes](https://numpy.org/doc/stable/release/1.24.0-notes.html#expired-deprecations) and...
