About hyperparameters

[rabiyaabbasi]

Hi, I am new to YOLOV5, so I have a couple of questions.

There are five hyp.yaml files in hyps directory (hyp.Objects365.yaml, hyp.scratch-low.yaml, hyp.scratch-high.yaml, hyp.scratch-med.yaml,hyp.VOC.yaml). My first question is what is the difference between all of them particularly hyp.scratch-low.yaml, hyp.scratch-high.yaml, and hyp.scratch-med.yaml. I know some values of parameters are changed but broadly whats the difference? And they will impact my model?

I am using the pre-trained model to train my custom data using command (!python train.py --img 416 --batch 20 --epochs 300 --data dataset.yaml --weights yolov5s.pt --cache). My question is why the default hyperparameter file for training is hyp.scratch-low.yaml? Can I change it to medium or high? If so how? Secondly, I have seen with hyperparameter evolution we can optimise the hyperparameters but if I change values of few hyperprameters in hyp.scratch-low.yaml file manually, will it bring changes to my model during training?

Thanks, your guidance is highly appreciated.

[glenn-jocher]

@rabiyaabbasi 👋 Hello! Thanks for asking about hyperparameters that define training and augmentation settings. Higher hyperparameters are used for larger models to delay overfitting.

YOLOv5 🚀 applies online imagespace and colorspace augmentations in the trainloader (but not the val_loader) to present a new and unique augmented Mosaic (original image + 3 random images) each time an image is loaded for training. Images are never presented twice in the same way.

Augmentation Hyperparameters

The hyperparameters used to define these augmentations are in your hyperparameter file (default data/hyp.scratch.yaml) defined when training:

python train.py --hyp hyp.scratch-low.yaml

yolov5/data/hyps/hyp.scratch-low.yaml

Lines 6 to 34 in b94b59e

	lr0: 0.01 # initial learning rate (SGD=1E-2, Adam=1E-3)
	lrf: 0.01 # final OneCycleLR learning rate (lr0 * lrf)
	momentum: 0.937 # SGD momentum/Adam beta1
	weight_decay: 0.0005 # optimizer weight decay 5e-4
	warmup_epochs: 3.0 # warmup epochs (fractions ok)
	warmup_momentum: 0.8 # warmup initial momentum
	warmup_bias_lr: 0.1 # warmup initial bias lr
	box: 0.05 # box loss gain
	cls: 0.5 # cls loss gain
	cls_pw: 1.0 # cls BCELoss positive_weight
	obj: 1.0 # obj loss gain (scale with pixels)
	obj_pw: 1.0 # obj BCELoss positive_weight
	iou_t: 0.20 # IoU training threshold
	anchor_t: 4.0 # anchor-multiple threshold
	# anchors: 3 # anchors per output layer (0 to ignore)
	fl_gamma: 0.0 # focal loss gamma (efficientDet default gamma=1.5)
	hsv_h: 0.015 # image HSV-Hue augmentation (fraction)
	hsv_s: 0.7 # image HSV-Saturation augmentation (fraction)
	hsv_v: 0.4 # image HSV-Value augmentation (fraction)
	degrees: 0.0 # image rotation (+/- deg)
	translate: 0.1 # image translation (+/- fraction)
	scale: 0.5 # image scale (+/- gain)
	shear: 0.0 # image shear (+/- deg)
	perspective: 0.0 # image perspective (+/- fraction), range 0-0.001
	flipud: 0.0 # image flip up-down (probability)
	fliplr: 0.5 # image flip left-right (probability)
	mosaic: 1.0 # image mosaic (probability)
	mixup: 0.0 # image mixup (probability)
	copy_paste: 0.0 # segment copy-paste (probability)

Augmentation Previews

You can view the effect of your augmentation policy in your train_batch*.jpg images once training starts. These images will be in your train logging directory, typically yolov5/runs/train/exp:

train_batch0.jpg shows train batch 0 mosaics and labels:

YOLOv5 Albumentations Integration

YOLOv5 🚀 is now fully integrated with Albumentations, a popular open-source image augmentation package. Now you can train the world's best Vision AI models even better with custom Albumentations 😃!

PR #3882 implements this integration, which will automatically apply Albumentations transforms during YOLOv5 training if albumentations>=1.0.3 is installed in your environment. See #3882 for full details.

Example train_batch0.jpg on COCO128 dataset with Blur, MedianBlur and ToGray. See the YOLOv5 Notebooks to reproduce:

Good luck 🍀 and let us know if you have any other questions!

[akashAD98]

@glenn-jocher its possible to apply augmention on specific class? i have 70 classes & i need to apply few augmentation on only specific to that class only, i dont want to apply all augmentions to all data

[25benjaminli]

This is a great idea.

[Abhiramborige]

thanks for the similar question.
why the hyp.scratch-low.yaml has the values (cls and obj) high than the values present in hyp.scratch-high.yaml
And also how should i set the values if I am in need of training on my custom dataset.
I wanted to detect small icons on a large images (80px size object on 2500px background)

PS:
hyp.scratch-low.yaml contains:
box: 0.05 # box loss gain
cls: 0.5 # cls loss gain
cls_pw: 1.0 # cls BCELoss positive_weight
obj: 1.0 # obj loss gain (scale with pixels)
iou_t: 0.20 # IoU training threshold

hyp.scratch-high.yaml contains:
box: 0.05 # box loss gain
cls: 0.3 # cls loss gain
cls_pw: 1.0 # cls BCELoss positive_weight
obj: 0.7 # obj loss gain (scale with pixels)
obj_pw: 1.0 # obj BCELoss positive_weight
iou_t: 0.20 # IoU training threshold

Thankyou

[Seema1997]

Hi @glenn-jocher Thanks for asking about hyperparameters that define training and augmentation settings. Higher hyperparameters are used for larger models to delay overfitting.

YOLOv5 🚀 applies online imagespace and colorspace augmentations in the trainloader (but not the val_loader) to present a new and unique augmented Mosaic (original image + 3 random images) each time an image is loaded for training. Images are never presented twice in the same way.

This is the reply to one of the above-asked questions. My question is whether this answer is also valid for YOLOv7, YOLOv8, YOLOv9, and YOLOv10. Can you please give some highlights on the data augmentation technique used for YOLOv7, YOLOv8, YOLOv9, and YOLOv10?