search

microsoft / fastmri-plus

Data labels and scripts for fastMRI.org
MIT License
72 stars 14 forks source link

About the application of this data set in target detection #5

Closed Breeze-Zero closed 2 years ago

Breeze-Zero commented 2 years ago

Hello, I have tried to convert the fast MRI Brain data set into PNG images in COCO format (I'm sure I flipped the images to align them with labels), and there are more than 3000 valid data.Then I called the open source target detection model for training, but the actual result was not ideal, the mAP could not exceed 0.1. I did some parameter experiments, but the result was not improved so far, so I began to doubt the quality and noise of the data set, and I hope to get confirmation from you

tianweiy commented 2 years ago

@834799106 did you manage to get higher number ?

Breeze-Zero commented 2 years ago

@834799106 did you manage to get higher number ?

Yes, compared to the accuracy of the same model in the general domain, I would have guessed that the accuracy would be lower, but it would be too low. In addition, I would like to ask whether you have tried this aspect and what the baseline results are for reference.

tianweiy commented 2 years ago

it is also super low in my case even if I only want to do classification

JiahaoHuang99 commented 1 year ago

Hi @tianweiy @Breeze-Zero ,

Have you solved this problem? Recently I tried YOLOv5 and YOLOv8 on the knee dataset but it is very low.