Closed ann1988425 closed 2 months ago
Hello, this step to build image pairs for model training can take a few hours depending on the specs of your computer. Is the software outputting training tiles? Try checking in the folder "pthDL", then a subfolder 'training\big_tiles\'. If that subfolder contains jpg files, the code is still running to build training and validation images.
Thank you for your kind reply, I have made that step and sucessfully trained my model. But I come up with a new problem when running test_model_performance (line 134 in train_image_segmentation_lung.m). The bug shows like:
And I trace back found that the tiff files in both train and test label file are all black, maybe that's the reason?
@ashleylk @ann1988425 ,Hi dear all,I want to know how you train the model,I followed the steps(https://labs.pathology.jhu.edu/kiemen/wp-content/uploads/sites/39/2023/12/Instructions-for-applying-CODA.pdf ,download data from https://drive.google.com/drive/folders/1K-wY_ArVGbEhebQD4AjOeERwx6-4Fw3G), but the generated TIF file cannot be opened in ImageScope. I have downloaded the latest version of ImageScope. just like this:
Then, the TIF file generated by downsampling with OpenSlide also encountered issues when annotating cell types. `import openslide from PIL import Image import numpy as np
slide = openslide.OpenSlide('lungs_001-.ndpi')
target_mpp = 1.0 #
current_mpp_x = float(slide.properties[openslide.PROPERTY_NAME_MPP_X]) current_mpp_y = float(slide.properties[openslide.PROPERTY_NAME_MPP_Y])
scale_factor_x = current_mpp_x / target_mpp scale_factor_y = current_mpp_y / target_mpp
level = slide.get_best_level_for_downsample(scale_factor_x) downsample = slide.level_downsamples[level]
# region = slide.read_region((0, 0), level, slide.level_dimensions[level])
region_rgb = region.convert("RGB")
region_np = np.array(region_rgb)
width, height = region.size scaled_width = int(width scale_factor_x) scaled_height = int(height scale_factor_y)
scaled_image = Image.fromarray(region_np).resize((scaled_width, scaled_height), Image.LANCZOS)
scaled_image.save('output_file.tiff', format='TIFF')
slide.close() ` just like this:
Could you please guide me on the steps for annotating cell types and training the model? Thank you very much.
@ashleylk @ann1988425 I also tried using the sample lung data provided by @ashleylk with CODA, but I encountered the same error as @ann1988425 and cannot proceed further.
If either of you knows a solution, could you please let me know?
Dear all,
Apologies for the error. The bug was that the testing image did not contain any annotations of metastases. There must be at least one annotation of each tissue layer in the testing dataset. I've updated the xml file for the testing image and re-uploaded it to the google drive. You may find the new file here: https://drive.google.com/drive/folders/1ztFE8Cv7U-DBEUaBmA2YcNjB8Ru-XbyK?usp=sharing
@ashleylk
Thank you very much. I will try again with the new file.
I would also like to make 3D reconstruction similar to the supplementary video in your paper. Is it possible to create this using CODA? I have been following the "Instructions for applying CODA" that you provided, but does it include instructions for 3D reconstruction?
@kubota119236541 Did you manage to build their lung dataset model ?
Hello, are your results trained using the 5 training sets and a test set provided by the sample annotations in your CODA sample dataset? Why does an infinite loop appear at this step when I use this training method?