Open synthdatagit opened 10 months ago
I made a workaround for the memory issue by resizing my original images.
Now another problem occurs.
When I press the yellow enter button on the web gui, nothing happens. Memory usage increases a bit but nothing is displayed..
Update, inspection output when pressing Enter
Uncaught TypeError: Cannot read properties of undefined (reading '0')
at World.setBorderColorImages (tsne.js:1484:14)
at Labels.updateLabels (tsne.js:3241:9)
at Labels.init (tsne.js:3094:8)
at Welcome.
Adding to @synthdatagit comments, I also got the yellow button having no action on the webpage.
The terminal reads normally but it does not load the images since no interaction is possible with the webpage. I was able to run the demo files (in the output folder) but, after fine-tuning with my own data, the images that I prepared cannot be run.
I also noticed that the 'image_vectors.npy' files that I produce only have (n_samples x 15) and not (n_samples x 2048) as it is meant to be according to:
In my example, the number of unique class_id is equal to 15. There might be a bug in how I am running the 'main.py' code (pixplotml > prep_pixplot_files). I am pretty sure that if I got 2048 arrays, I will be able to run 'pixplot.py' (within pixplotml > pixplot_server > pixplot) and generate the webpage without issues.
I hope you can help me see my data. Thanks!.
Cordially, Marco
Dear @alexhock,
I managed to figure out the solution to my issue above. The last stripped off layer of the model was being flattened and provided n_columns = number of classes in 'class_id'. The solution is to replace (in prep_pixplot_files > 'main.py') the following:
emb_model.fc = torch.nn.Identity(2048) emb_model = emb_model.to(device)
by
emb_model = torch.nn.Sequential(*list(emb_model.children())[:-1])
I hope this is helpful for someone else in the future.
Best wishes, Marco
Hi,
first of all. Really cool repo
I have tried to use it for my own data, but I encounter an issue with memory. My metadata.csv file contains approx. 32000 lines (32000 objects in 2000 images).
When I run pixplot.py I get the following error
Traceback (most recent call last): File "C:\work\pixplotml\pixplot_server\pixplot\pixplot.py", line 1677, in
parse()
File "C:\work\pixplotml\pixplot_server\pixplot\pixplot.py", line 1673, in parse
process_images(config)
File "C:\work\pixplotml\pixplot_server\pixplot\pixplot.py", line 220, in process_images
) = load_and_filter_images(kwargs)
File "C:\work\pixplotml\pixplot_server\pixplot\pixplot.py", line 286, in load_and_filter_images
all_loaded_images = load_input_files(kwargs)
File "C:\work\pixplotml\pixplot_server\pixplot\pixplot.py", line 277, in load_input_files
img = Image(path, {"metadata": metadata, "vec": vec})
File "C:\work\pixplotml\pixplot_server\pixplot\pixplot.py", line 1483, in init
self.original = load_image(self.path)
File "C:\work\pixplotml\pixplot_server\pixplot\pixplot.py", line 1477, in load_image
return pil_image.open(image_path).convert("RGB")
File "C:\Users\xxxxxx\Anaconda3\envs\pixplotml\lib\site-packages\PIL\Image.py", line 923, in convert
return self.copy()
File "C:\Users\xxxxxx\Anaconda3\envs\pixplotml\lib\site-packages\PIL\Image.py", line 1179, in copy
return self._new(self.im.copy())
MemoryError
I have attached the memory usage![memory](https://github.com/alexhock/pixplotml/assets/113902383/e3b4b2a4-a056-49b3-8aba-c5394c606264)
Do you have any suggestions to working with larger datasets in pixplotml ?