shensheng272 comments

Results 43 comments of


                                            shensheng272

Allow images with upper case file extensions. e.g. foo.JPG

Thanks for pr. But currently IMG_FORMAT is used with `.lower()` . So images with upper case file extensions will cause no trouble.

Allow images with upper case file extensions. e.g. foo.JPG

> @shensheng272 There is a bug [here](https://github.com/meituan/YOLOv6/blob/main/yolov6/data/datasets.py#L572) where the code is not using `.lower()`. So either that line needs to be changed to call `.lower()`, or `IMG_FORMAT` needs to include...

We need to have Confusion matrix, Precision, Recall and F1 score

Latest code will print conf on best mF1 and print P R under iou50, when eval model. Also default plot curves like yolov5. Also you can log infos you want...

Would be great to add Per-Class "Precision, Recall, mAPs" to the Evaluation Metrics | 请考虑对每一类加上验证指标，如Precision，Recall，mAPs

Latest code will print conf on best mF1 and print P R under iou50. Also default plot curves like yolov5. Also you can log infos you want near https://github.com/meituan/YOLOv6/blob/main/yolov6/core/evaler.py#L188

Custom_Dataset_ error - assert osp.exists(label_dir), f"{label_dir} is an invalid directory path!"

you might check your ymal to follow ``` is_coco: False names: ['Sign'] nc: 1 train: ./images/train val: ./images/val ```

为什么最新的拉下来训练的很慢？比半个月拉下来额慢了好几本配置一样

> 请问可以实现训练中控制不用OTA么 V0.2.0 版本去掉了OTA来加速训练。欢迎使用。

为什么最新的拉下来训练的很慢？比半个月拉下来额慢了好几本配置一样

是不是v5版本里面没有开mosaic，或者mosaic等一些增强的概率不同？一些预处理是比较耗时的。

How to reproduce the results

New models and scripts for reproducing train/val result are available now. More info in [readme](https://github.com/meituan/YOLOv6/blob/main/README.md). By the way, we publish result on different test scheme comparing to yolov5/yolov7. Still you...

Model training and evaluation problems

Multiple train/val/test dirs

For now, we merge [PR](https://github.com/meituan/YOLOv6/pull/377) about recursive train/val folder, might alleviate the problem.