LSH9832 / edgeyolo

an edge-real-time anchor-free object detector with decent performance
Apache License 2.0
442 stars 59 forks source link

Training in Yolo format #18

Open usama-x930 opened 1 year ago

usama-x930 commented 1 year ago

I have YOLO Annotated text dataset in Folder structure train: images/.jpg labels/.txt

val: images/.jpg labels/.txt

Got this Error after 1 epoch:

Loading and preparing results... 20230314_012420 edgeyolo.train.trainer:480 - error: Results do not correspond to current coco set

LSH9832 commented 1 year ago

it means there's at least one image with empty label in your val dataset, just check your txt files and remove the empty files, then set use_cache to false to re-generate cache.

ramonhollands commented 1 year ago

I got the same error and discovered that the problem in my case was caused because of different image formats mingled. By default, the Yolo dataloader takes only the '.jpg' images, while all annotations are taken (also from other image formats, like '.png'). That causes a mismatch.

I created a fix to take all images (despite of their format) and after testing, Ill create a pull request.

LSH9832 commented 1 year ago

I got the same error and discovered that the problem in my case was caused because of different image formats mingled. By default, the Yolo dataloader takes only the '.jpg' images, while all annotations are taken (also from other image formats, like '.png'). That causes a mismatch.

I created a fix to take all images (despite of their format) and after testing, Ill create a pull request.

Well, typically the first step before annotating a dataset is to unify the image format, so I think it's the problem of nonstandard process when creating the dataset, not the code