Chilicyy comments

Results 107 comments of


                                            Chilicyy

What does num_layers mean and how to deal with a feature extractor that has 4 feature pyramid outputs?

@sarmientoj24 You can try to add some operations like RepBlocks+1*1conv+upsample to build P2 outputs . Modify the code in ./yolov6/models/reppan.py.

Invitation to Meituan YOLOv6 team to speak at YOLO VISION event on September 27th, 2022

Thanks for your invitation. It sounds interesting, and we will contact you by email for more details.

Adding Support for other Dataset formats

Of course you can, but for the sake of simplicity and readability of the code and consistent format, we recommend adding some of the more commonly used detection dataset formats....

Adding Support for other Dataset formats

It sounds great. Welcome to contribute!

Is there a way to do half precision training?

Currently we support mixed precision training by default, which means using fp16 during fast-forward inference, and using fp32 during the back-propagation process.

Need help in utilizing 4 feature pyramids to create 3 feature maps

@sarmientoj24 You can try to add some operations like RepBlocks+1*1conv+upsample to build P2 outputs . Modify the code in ./yolov6/models/reppan.py.

yolov6 nano在训练的最后100个epoch有特殊操作吗?

您好，small与tiny的区别除了您说的这些之外没有其他了，一般建议使用8卡batchsize为256的训练配置，谢谢！

training time

@johnnysclai Thanks for your suggestion. We will consider it and add this feature ASAP.

mAP is low and no detection result.

hi @sssssshf If dataset has been trained with yolov5/7 or older version of yolov6, the dataset cache might be incompatible for yolov6. Add --check-images and --check-images to create new cache.

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

@JiaPai12138 您好，我们这边分别在COCO和VOC数据集上对比了yolov5和yolov6的训练时间，两者训练时长相差不大。看到你这边反馈的情况是同样的数据集yolov5一轮训练时长为1分钟而v6要3.5～4分钟。请问下在v5和v6上用的batchsize是一致的吗？