Does anyone know how to process the data?

https://github.com/AntiAegis/pytorch-image-models is private repo. I cannot clone the repo. Therefore, I am not able to run the scripts.

should add: timm==0.1.12

When I am running this command: python inference_video.py --watch --use_cuda --checkpoint path_to_checkpoint/model_best.pth
I get this error. How do I resolve it?

I am having this issue when running training from scratch

Traceback (most recent call last):
  File "train.py", line 101, in <module>
    main(config, args.resume)
  File "train.py", line 55, in main
    trainer.train()
  File "/path/Human-Segmentation-PyTorch/base/base_trainer.py", line 95, in train
    result = self._train_epoch(epoch)
  File "/path/Human-Segmentation-PyTorch/trainer/trainer.py", line 81, in _train_epoch
    loss = self.loss(output, target)
  File "/path/Human-Segmentation-PyTorch/evaluation/losses.py", line 18, in dice_loss
    targets = torch.zeros_like(logits).scatter_(dim=1, index=targets.type(torch.int64), src=torch.tensor(1.0))
RuntimeError: Index tensor must have the same number of dimensions as src tensor

I guess I might have an issue in the dataset labeling. What is the correct format? I used an image of the same size for the original image and have 1 channel (I have 2 classes), so that is a mask of 1 channel containing 0 and 1 (or 0 and 255)

/Human-Segmentation-PyTorch$ python inference_video.py --watch --checkpoint ./checkpoint/UNet_ResNet18.pth
Traceback (most recent call last):
File "inference_video.py", line 80, in
model.load_state_dict(trained_dict, strict=False)
File "/home/anaconda3/envs/humansegmentation/lib/python3.6/site-packages/torch/nn/modules/module.py", line 845, in load_state_dict
self.class.name, "\n\t".join(error_msgs)))
RuntimeError: Error(s) in loading state_dict for UNet:
size mismatch for decoder1.deconv.weight: copying a param with shape torch.Size([512, 256, 4, 4]) from checkpoint, the shape in current model is torch.Size([1280, 96, 4, 4]).
size mismatch for decoder1.deconv.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([96]).
size mismatch for decoder2.deconv.weight: copying a param with shape torch.Size([256, 128, 4, 4]) from checkpoint, the shape in current model is torch.Size([96, 32, 4, 4]).
size mismatch for decoder2.deconv.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([32]).
size mismatch for decoder3.deconv.weight: copying a param with shape torch.Size([128, 64, 4, 4]) from checkpoint, the shape in current model is torch.Size([32, 24, 4, 4]).
size mismatch for decoder3.deconv.bias: copying a param with shape torch.Size([64]) from checkpoint, the shape in current model is torch.Size([24]).
size mismatch for decoder4.deconv.weight: copying a param with shape torch.Size([64, 64, 4, 4]) from checkpoint, the shape in current model is torch.Size([24, 16, 4, 4]).
size mismatch for decoder4.deconv.bias: copying a param with shape torch.Size([64]) from checkpoint, the shape in current model is torch.Size([16]).
size mismatch for conv_last.0.weight: copying a param with shape torch.Size([3, 64, 3, 3]) from checkpoint, the shape in current model is torch.Size([3, 16, 3, 3]).

Hi~ Thanks for your wonderful job.
I want to use your code to train but the download speed of Supervisely dataset(from supervise.ly) is very slow.
How you download this dataset or do you have a googledrive link of this dataset?
Thanks~

Hi, thanks for your nice work and enthusiasm of open-source.

We are members of MMSegmentation, its aim is to support more models and datasets for convenient usage by communities. Because of limited human sources, we hope communities could also join us. Would you like to support this Portrait Segmentation dataset like this customization pipeline, which would also in turn enlarge the popularity of your excellent work?

Best,

Is there any way to get bbox values?

Thank you for your work.What is the meaning of the input_sz parameter? Why do you want to do padding=0?

dear,
~/Human-Segmentation-PyTorch$ python inference_video.py --watch --checkpoint ./checkpoint/UNet_ResNet18.pth
OpenCV: FFMPEG: tag 0x5634504d/'MP4V' is not supported with codec id 12 and format 'mp4 / MP4 (MPEG-4 Part 14)'
OpenCV: FFMPEG: fallback to use tag 0x7634706d/'mp4v'
Traceback (most recent call last):
File "inference_video.py", line 80, in
model.load_state_dict(trained_dict, strict=False)
File "/home/anaconda3/envs/humansegmentation/lib/python3.6/site-packages/torch/nn/modules/module.py", line 845, in load_state_dict
self.class.name, "\n\t".join(error_msgs)))
RuntimeError: Error(s) in loading state_dict for UNet:
size mismatch for decoder1.deconv.weight: copying a param with shape torch.Size([512, 256, 4, 4]) from checkpoint, the shape in current model is torch.Size([1280, 96, 4, 4]).
size mismatch for decoder1.deconv.bias: copying a param with shape torch.Size([256]) from checkpoint, the shape in current model is torch.Size([96]).
size mismatch for decoder2.deconv.weight: copying a param with shape torch.Size([256, 128, 4, 4]) from checkpoint, the shape in current model is torch.Size([96, 32, 4, 4]).
size mismatch for decoder2.deconv.bias: copying a param with shape torch.Size([128]) from checkpoint, the shape in current model is torch.Size([32]).
size mismatch for decoder3.deconv.weight: copying a param with shape torch.Size([128, 64, 4, 4]) from checkpoint, the shape in current model is torch.Size([32, 24, 4, 4]).
size mismatch for decoder3.deconv.bias: copying a param with shape torch.Size([64]) from checkpoint, the shape in current model is torch.Size([24]).
size mismatch for decoder4.deconv.weight: copying a param with shape torch.Size([64, 64, 4, 4]) from checkpoint, the shape in current model is torch.Size([24, 16, 4, 4]).
size mismatch for decoder4.deconv.bias: copying a param with shape torch.Size([64]) from checkpoint, the shape in current model is torch.Size([16]).
size mismatch for conv_last.0.weight: copying a param with shape torch.Size([3, 64, 3, 3]) from checkpoint, the shape in current model is torch.Size([3, 16, 3, 3]).

Can you show me what is error?

Hi, thank you for your great work. I downloaded your model on Google dirver and found that the model size is much larger than the model size written in readme. Is this normal?

hello, thanks for your work. Could you provide google drive url for your pretrained models? thanks

Video portrait segmentation is not unstable， often flickering when use unet model，and loss used crossentropy. i am wonder I what causes the flicker and whether it is related to training loss。

I am looking forward to answer。

get such error,
installed timm using pip3 install timm

run python3 inference_webcam.py --checkpoint UNet_MobileNetV2.pth

which paper is this project reimplementing? Thanks.

when i run train.py unet_config
after trained 1 epoch,error shows:

Traceback
train.py line 101
trainer.train()
base_trainer.py line95
result = self.train_epoch(epoch)
trainer/trainer.py line98
self.writer_train.add_image('train/output',make_grid(.....))
torchvision/utils.py ,line66, in make_grid
norm_range(tensor,range)
torchvision/utils.py ,line60, in norm_range
norm_ip(t,t.min,max=max)
torchvision/utils.py ,line53, in norm_ip
img.clamp_(argument 'min' must be Number ,not tensor)

can you help me ,my own dataset input is rgb ,mask is 1 channel (0,1)

There is an error when I want to download the pretrained weights!

if any one can provide a download link?

HI, have you done this project in VOC2012 datasets? Something is wrong with this datasets.

Thanks for you work!
I have two questions:

• Where to get the "ResNet" backbones pretrained weights?

•  def resnet50(pretrained=None, **kwargs):
    model = ResNet(Bottleneck, [3, 4, 6, 3], **kwargs)
    if pretrained is not None:
    	model._load_pretrained_model(pretrained)
    return model

  where is the definition of function _load_pretrained_model?

"Page not found" showed on GIthub.

Hi, I just wonder where can i get the pretrained models, whose paths in the json files are like:
"pretrained_backbone": "/data4/livesegmentation/thuync/PyTorch-pretrained/resnet18.pth"
or
"pretrained_backbone": "/root/pretrain/resnet18.pth"
or
"/workspace/pretrain/resnet18.pth"?

After ran the command
$ git submodule update --init --recursive

It stuck without response. I checked the repo that "pytorch-image-models @ 2be1fd0 " pointed to a webpage with 404 error( not found).

Thanks for your great work!
In the table at the bottom of readme, results of different models are showed, and I've noticed that Unet-Mobilenet get the best iou rather than models that have much more parameters. What do you think have caused this? Thanks!

Hello. Thank you very much for sharing the code and model.

What is the license for this project and the pre-training model? Can I use it in commercial projects?

I really need your reply.

from timm.models.resnet import default_cfgs, load_pretrained, BasicBlock, Bottleneck
ImportError: cannot import name 'load_pretrained'

and in timm.models.resnet , there is no ‘load_pretrained’
Can someone tell me how to do

How to run the demo on a still image? I think its provided from webcam and camera. Thanks

dear,
Where can I download model_best.pth?

Could you give me some instructions about how to calculate the mIOU with your code?

when i load the trained model UNet_ResNet18.pth with torch.load(DeepLabV3Plus_ResNet18.pth). a error had occur,the detail as follows.

i will appreciate it if you can provide some advices. thanks a lot