Question about CVC-ClinicDB dataset

med-air / Endo-FM

[MICCAI'23] Foundation Model for Endoscopy Video Analysis via Large-scale Self-supervised Pre-train

Apache License 2.0

154 stars 15 forks source link

Question about CVC-ClinicDB dataset #17

Closed zjuxiaobaiq closed 3 months ago

zjuxiaobaiq commented 4 months ago

Hi, Great work! Can you help me with some questions? I wonder that why you need to do normalization firstly and then transfer the pix value to 0-255. As i find that the min picture pix value is 0 and max picture pix value is 255. I also wonder that tif picture only has one channel, but it can be displayed in color. And after cv2 reads the tif picture, i cannot save the tif picture in color anymore.

Kyfafyd commented 4 months ago

Hi @zjuxiaobaiq , thanks for your interest!

I remember the original pix value is not in 0-255, so it is normalized first and then transfer to 0-255.
This function Image.fromarray requires the pix value 0-255.
From my personal side, it is not necessary to save tif image in color, as it is in grey style.

zjuxiaobaiq commented 4 months ago

Thank you for your reply. Actually, the tif image is in "rgb" style. I find that using cv2 to read tif image results in one channel while using others do not. A new question is that reading CVC-ClinicDB data in gray style is the common use? What are the benefits of reading CVC-ClinicDB data in gray style？

Kyfafyd commented 4 months ago

Thanks for your information! Interesting~ I am not very sure if this is common usage. You can try to train the model with 3-channel and compare with the one trained with 1-channel.