Lego dataset win_size exceeds error

[engrkhurramshabbir]

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
File "/home/khurram/kaolin-wisp/app/nerf/main_nerf.py", line 488, in
app.run() # Run in interactive mode
File "/home/khurram/kaolin-wisp/wisp/renderer/app/wisp_app.py", line 253, in run
app.run() # App clock should always run as frequently as possible (background tasks should not be limited)
File "/home/khurram/anaconda3/envs/wisp/lib/python3.9/site-packages/glumpy/app/init.py", line 362, in run
run(duration, framecount)
File "/home/khurram/anaconda3/envs/wisp/lib/python3.9/site-packages/glumpy/app/init.py", line 344, in run
count = backend.process(dt)
File "/home/khurram/anaconda3/envs/wisp/lib/python3.9/site-packages/glumpy/app/window/backends/backend_glfw_imgui.py", line 451, in process
window.dispatch_event('on_idle', dt)
File "/home/khurram/anaconda3/envs/wisp/lib/python3.9/site-packages/glumpy/app/window/event.py", line 396, in dispatch_event
if getattr(self, event_type)(*args):
File "/home/khurram/kaolin-wisp/wisp/renderer/app/wisp_app.py", line 521, in _run_hook
hook()
File "/home/khurram/kaolin-wisp/wisp/trainers/base_trainer.py", line 333, in iterate
self.end_epoch()
File "/home/khurram/kaolin-wisp/wisp/trainers/base_trainer.py", line 294, in end_epoch
self.validate()
File "/home/khurram/kaolin-wisp/wisp/trainers/multiview_trainer.py", line 225, in validate
evaluation_results = self.evaluate_metrics(self.validation_dataset, lods[-1],
File "/home/khurram/kaolin-wisp/wisp/trainers/multiview_trainer.py", line 118, in evaluate_metrics
ssim_total += ssim(rb.rgb[...,:3], gts[...,:3])
File "/home/khurram/kaolin-wisp/wisp/ops/image/metrics.py", line 84, in ssim
return skimage.metrics.structural_similarity(
File "/home/khurram/anaconda3/envs/wisp/lib/python3.9/site-packages/skimage/metrics/_structural_similarity.py", line 178, in structural_similarity
raise ValueError(
ValueError: win_size exceeds image extent. Either ensure that your images are at least 7x7; or pass win_size explicitly in the function call, with an odd value less than or equal to the smaller side of your images. If your images are multichannel (with color channels), set channel_axis to the axis number corresponding to the channels.

[orperel]

Hi @engrkhurramshabbir , SSIM seems to work fine on my end, which configuration are you running?

[engrkhurramshabbir]

I am using following configuration
cd kaolin-wisp
python3 app/nerf/main_nerf.py --config app/nerf/configs/nerf_octree.yaml --dataset-path /path/to/lego

[engrkhurramshabbir]

My data structure is as follows:
test
train
val
transform_test.json
transform_train.json
transform_val.json

[engrkhurramshabbir]

I even tried multiply configurations and other data e.g. fox. But still getting same error. Problem seems to define win_size explicitly for the given data set in metrics.py, multiview_trainer.py and base_trainer.py

File "/home/khurram/kaolin-wisp/wisp/renderer/app/wisp_app.py", line 521, in _run_hook
hook()
File "/home/khurram/kaolin-wisp/wisp/trainers/base_trainer.py", line 333, in iterate
self.end_epoch()
File "/home/khurram/kaolin-wisp/wisp/trainers/base_trainer.py", line 294, in end_epoch
self.validate()
File "/home/khurram/kaolin-wisp/wisp/trainers/multiview_trainer.py", line 225, in validate
evaluation_results = self.evaluate_metrics(self.validation_dataset, lods[-1],
File "/home/khurram/kaolin-wisp/wisp/trainers/multiview_trainer.py", line 118, in evaluate_metrics
ssim_total += ssim(rb.rgb[...,:3], gts[...,:3])
File "/home/khurram/kaolin-wisp/wisp/ops/image/metrics.py", line 84, in ssim
return skimage.metrics.structural_similarity(
File "/home/khurram/anaconda3/envs/wisp/lib/python3.9/site-packages/skimage/metrics/_structural_similarity.py", line 178, in structural_similarity
raise ValueError(
ValueError: win_size exceeds image extent. Either ensure that your images are at least 7x7; or pass win_size explicitly in the function call, with an odd value less than or equal to the smaller side of your images. If your images are multichannel (with color channels), set channel_axis to the axis number corresponding to the channels.

[jaeheungs]

I think it may be an issue with skimage version

I change the file at wisp/ops/image/metrics.py to include the channel_axis, namely:

    return skimage.metrics.structural_similarity(
        rgb[..., :3].cpu().numpy(),
        gts[..., :3].cpu().numpy(),
        multichannel=True,
        data_range=1,
        gaussian_weights=True,
        sigma=1.5,
        channel_axis=-1)

and it works now.