xwying / torchshow Goto Github PK

Hi there, I am currently visualising feature maps in yolov7 for each layer and I was wondering if it was possible for each image to have a subtitle? for example "Feature map 1"

Thank you so much for your project! It really saves me time in visualizing PyTorch tensors. Do you have any plan to show a single video in a grid layout like this?

Hi, recently i meet this bug when saving mask, but this bug is a little bit strange because it not always happen but for certain images‘ gt mask,

File "/home/zmz/Mask2Former/mask2former/maskformer_model.py", line 316, in forward
torchshow.save(masks[0]>0,"gt_mask.jpg")
File "/home/zmz/miniconda3/envs/py38/lib/python3.8/site-packages/torchshow/torchshow.py", line 19, in save
show(x, save=True, file_path=path, **kwargs)
File "/home/zmz/miniconda3/envs/py38/lib/python3.8/site-packages/torchshow/torchshow.py", line 97, in show
display_plt(plot_list, **kwargs)
File "/home/zmz/miniconda3/envs/py38/lib/python3.8/site-packages/torchshow/visualization.py", line 156, in display_plt
fig.savefig(file_path, bbox_inches = 'tight', pad_inches=0)
File "/home/zmz/miniconda3/envs/py38/lib/python3.8/site-packages/matplotlib/figure.py", line 3046, in savefig
self.canvas.print_figure(fname, **kwargs)
File "/home/zmz/miniconda3/envs/py38/lib/python3.8/site-packages/matplotlib/backend_bases.py", line 2295, in print_figure
self.figure.draw(renderer)
File "/home/zmz/miniconda3/envs/py38/lib/python3.8/site-packages/matplotlib/artist.py", line 73, in draw_wrapper
result = draw(artist, renderer, *args, **kwargs)
File "/home/zmz/miniconda3/envs/py38/lib/python3.8/site-packages/matplotlib/artist.py", line 50, in draw_wrapper
return draw(artist, renderer)
File "/home/zmz/miniconda3/envs/py38/lib/python3.8/site-packages/matplotlib/figure.py", line 2837, in draw
mimage._draw_list_compositing_images(
File "/home/zmz/miniconda3/envs/py38/lib/python3.8/site-packages/matplotlib/image.py", line 132, in _draw_list_compositing_images
a.draw(renderer)
File "/home/zmz/miniconda3/envs/py38/lib/python3.8/site-packages/matplotlib/artist.py", line 50, in draw_wrapper
return draw(artist, renderer)
File "/home/zmz/miniconda3/envs/py38/lib/python3.8/site-packages/matplotlib/axes/_base.py", line 3091, in draw
mimage._draw_list_compositing_images(
File "/home/zmz/miniconda3/envs/py38/lib/python3.8/site-packages/matplotlib/image.py", line 132, in _draw_list_compositing_images
a.draw(renderer)
File "/home/zmz/miniconda3/envs/py38/lib/python3.8/site-packages/matplotlib/artist.py", line 50, in draw_wrapper
return draw(artist, renderer)
File "/home/zmz/miniconda3/envs/py38/lib/python3.8/site-packages/matplotlib/image.py", line 646, in draw
im, l, b, trans = self.make_image(
File "/home/zmz/miniconda3/envs/py38/lib/python3.8/site-packages/matplotlib/image.py", line 956, in make_image
return self._make_image(self._A, bbox, transformed_bbox, clip,
File "/home/zmz/miniconda3/envs/py38/lib/python3.8/site-packages/matplotlib/image.py", line 555, in _make_image
output_alpha = _resample( # resample alpha channel
File "/home/zmz/miniconda3/envs/py38/lib/python3.8/site-packages/matplotlib/image.py", line 193, in _resample
_image.resample(data, out, transform,
ValueError: Unsupported dtype

torchshow.save(images)
Traceback (most recent call last):
  File "torchshow/utils.py", line 60, in tensor_to_array
    return x.detach().clone().cpu().numpy()
TypeError: Got unsupported ScalarType BFloat16

BFloat16 is currently not supported. It would be preferable to optimize this if possible.

Hi,
I get this error when I try to import torchshow

torchshow.set_image_mean([0., 0., 0.])
torchshow.set_image_std([1., 1., 1.])
torchshow.show(img)


---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)

File /somepath/python3.8/site-packages/torchshow/visualization.py:289, in vis_image(x, unnormalize, **kwargs)
    [287]     if user_std == None:
    [288]         user_std = [1.] * x.shape[-1] # Initialize std to 1 if not specified.
--> [289]     x = unnormalize(x, user_mean, user_std)
    [291] elif unnormalize=='auto':
    [292]     x = auto_unnormalize_image(x)

TypeError: 'str' object is not callable

It seems to be caused by the variable name and method name being the same 'unnormalize'.
And this case should be added to the test.

How to show with size?
ex, in matplotlib , I config plt.figure(figsize=(10,10))

Currently, it seems that for input pytorch float tensor between -1 and 1, a min-max normalization is taken:

However, usually people use transforms.Normalize((0.5,), (0.5,)) to normalize image.
So I think when x is between -1 and 1, it should be x * 0.5 + 0.5 to keep the image color consistent.

I know it can be achieved by setting set_image_mean([0.5, 0.5, 0.5]) and set_image_std([0.5, 0.5, 0.5]), but this may be used for other special cases. For the most common cases, i think we should adopt the above method.

Any idea?

I have a relatively large tensor I'd like to visualize, multiple iterations of a 2048x3292 tensor. When I use save on a list of 4 of these tensors, the image gets shrunk down to 609x87, which is way too small to see the details I need to see. Is there an option available to make the save function not compress the image in size? I tried ncols and nrows, but it still shrunk it.

>>> print(len(tensors), tensors[0].shape)
4 torch.Size([2048, 3292])
>>> ts.save([tensors], args.tensor_name + ".png", nrows=2048, ncols=3292)

--> result is a 609x87 image

Also, as an aside, is there a convenient way to map positive numbers to green and negative numbers to red (or any other color scheme), or should I handle remapping those values before passing to torchshow?

Thanks!

Hi @xwying ,

Is there a speed difference compared to other libraries? What are the advantages?

I try to use this command in debug console with VScode jupyter:

ts.show(random_batch[0])

but return None and no image window pops up

I'm visualizing my feature map using torchshow. It's very useful to me. However, torchshow only support grap color map when visualizing tensor whose shape is (N, C, H, W), where C>=1. I need color map like cv2.COLORMAP_JET to show clear information about feature tensors, which will make things easier, thanks!

Hi, i meet this warning when i use torchshow to save masks in 'for' loop. Is there anyway to solve this problem?

envs/py38/lib/python3.8/site-packages/torchshow/visualization.py:108: RuntimeWarning: More than 20 figures have been opened. Figures created through the pyplot interface (matplotlib.pyplot.figure) are retained until explicitly closed and may consume too much memory. (To control this warning, see the rcParam figure.max_open_warning).
fig, axes = plt.subplots(nrows=nrows, ncols=ncols, squeeze=False)