Test-Time Augmentation (TTA) Tutorial

[glenn-jocher]

📚 This guide explains how to use Test Time Augmentation (TTA) during testing and inference for improved mAP and Recall with YOLOv5 🚀. UPDATED 25 September 2022.

Before You Start

Clone repo and install requirements.txt in a Python>=3.7.0 environment, including PyTorch>=1.7. Models and datasets download automatically from the latest YOLOv5 release.

git clone https://github.com/ultralytics/yolov5  # clone
cd yolov5
pip install -r requirements.txt  # install

Test Normally

Before trying TTA we want to establish a baseline performance to compare to. This command tests YOLOv5x on COCO val2017 at image size 640 pixels. yolov5x.pt is the largest and most accurate model available. Other options are yolov5s.pt, yolov5m.pt and yolov5l.pt, or you own checkpoint from training a custom dataset ./weights/best.pt. For details on all available models please see our README table.

$ python val.py --weights yolov5x.pt --data coco.yaml --img 640 --half

Output:

val: data=./data/coco.yaml, weights=['yolov5x.pt'], batch_size=32, imgsz=640, conf_thres=0.001, iou_thres=0.65, task=val, device=, single_cls=False, augment=False, verbose=False, save_txt=False, save_hybrid=False, save_conf=False, save_json=True, project=runs/val, name=exp, exist_ok=False, half=True
YOLOv5 🚀 v5.0-267-g6a3ee7c torch 1.9.0+cu102 CUDA:0 (Tesla P100-PCIE-16GB, 16280.875MB)

Fusing layers... 
Model Summary: 476 layers, 87730285 parameters, 0 gradients

val: Scanning '../datasets/coco/val2017' images and labels...4952 found, 48 missing, 0 empty, 0 corrupted: 100% 5000/5000 [00:01<00:00, 2846.03it/s]
val: New cache created: ../datasets/coco/val2017.cache
               Class     Images     Labels          P          R     mAP@.5 mAP@.5:.95: 100% 157/157 [02:30<00:00,  1.05it/s]
                 all       5000      36335      0.746      0.626       0.68       0.49
Speed: 0.1ms pre-process, 22.4ms inference, 1.4ms NMS per image at shape (32, 3, 640, 640)  # <--- baseline speed

Evaluating pycocotools mAP... saving runs/val/exp/yolov5x_predictions.json...
...
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.504  # <--- baseline mAP
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.688
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.546
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.351
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.551
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.644
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.382
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.628
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.681  # <--- baseline mAR
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.524
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.735
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.826

Test with TTA

Append --augment to any existing val.py command to enable TTA, and increase the image size by about 30% for improved results. Note that inference with TTA enabled will typically take about 2-3X the time of normal inference as the images are being left-right flipped and processed at 3 different resolutions, with the outputs merged before NMS. Part of the speed decrease is simply due to larger image sizes (832 vs 640), while part is due to the actual TTA operations.

$ python val.py --weights yolov5x.pt --data coco.yaml --img 832 --augment --half

Output:

val: data=./data/coco.yaml, weights=['yolov5x.pt'], batch_size=32, imgsz=832, conf_thres=0.001, iou_thres=0.6, task=val, device=, single_cls=False, augment=True, verbose=False, save_txt=False, save_hybrid=False, save_conf=False, save_json=True, project=runs/val, name=exp, exist_ok=False, half=True
YOLOv5 🚀 v5.0-267-g6a3ee7c torch 1.9.0+cu102 CUDA:0 (Tesla P100-PCIE-16GB, 16280.875MB)

Fusing layers... 
/usr/local/lib/python3.7/dist-packages/torch/nn/functional.py:718: UserWarning: Named tensors and all their associated APIs are an experimental feature and subject to change. Please do not use them for anything important until they are released as stable. (Triggered internally at  /pytorch/c10/core/TensorImpl.h:1156.)
  return torch.max_pool2d(input, kernel_size, stride, padding, dilation, ceil_mode)
Model Summary: 476 layers, 87730285 parameters, 0 gradients
val: Scanning '../datasets/coco/val2017' images and labels...4952 found, 48 missing, 0 empty, 0 corrupted: 100% 5000/5000 [00:01<00:00, 2885.61it/s]
val: New cache created: ../datasets/coco/val2017.cache
               Class     Images     Labels          P          R     mAP@.5 mAP@.5:.95: 100% 157/157 [07:29<00:00,  2.86s/it]
                 all       5000      36335      0.718      0.656      0.695      0.503
Speed: 0.2ms pre-process, 80.6ms inference, 2.7ms NMS per image at shape (32, 3, 832, 832)  # <--- TTA speed

Evaluating pycocotools mAP... saving runs/val/exp2/yolov5x_predictions.json...
...
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.516  # <--- TTA mAP
 Average Precision  (AP) @[ IoU=0.50      | area=   all | maxDets=100 ] = 0.701
 Average Precision  (AP) @[ IoU=0.75      | area=   all | maxDets=100 ] = 0.562
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.361
 Average Precision  (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.564
 Average Precision  (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.656
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=  1 ] = 0.388
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets= 10 ] = 0.640
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=   all | maxDets=100 ] = 0.696  # <--- TTA mAR
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.553
 Average Recall     (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.744
 Average Recall     (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.833

Inference with TTA

detect.py TTA inference operates identically to val.py TTA: simply append --augment to any existing detect.py command:

$ python detect.py --weights yolov5s.pt --img 832 --source data/images --augment

Output:

detect: weights=['yolov5s.pt'], source=data/images, imgsz=832, conf_thres=0.25, iou_thres=0.45, max_det=1000, device=, view_img=False, save_txt=False, save_conf=False, save_crop=False, nosave=False, classes=None, agnostic_nms=False, augment=True, update=False, project=runs/detect, name=exp, exist_ok=False, line_thickness=3, hide_labels=False, hide_conf=False, half=False
YOLOv5 🚀 v5.0-267-g6a3ee7c torch 1.9.0+cu102 CUDA:0 (Tesla P100-PCIE-16GB, 16280.875MB)

Downloading https://github.com/ultralytics/yolov5/releases/download/v5.0/yolov5s.pt to yolov5s.pt...
100% 14.1M/14.1M [00:00<00:00, 81.9MB/s]

Fusing layers... 
Model Summary: 224 layers, 7266973 parameters, 0 gradients
image 1/2 /content/yolov5/data/images/bus.jpg: 832x640 4 persons, 1 bus, 1 fire hydrant, Done. (0.029s)
image 2/2 /content/yolov5/data/images/zidane.jpg: 480x832 3 persons, 3 ties, Done. (0.024s)
Results saved to runs/detect/exp
Done. (0.156s)

PyTorch Hub TTA

TTA is automatically integrated into all YOLOv5 PyTorch Hub models, and can be accessed by passing augment=True at inference time.

import torch

# Model
model = torch.hub.load('ultralytics/yolov5', 'yolov5s')  # or yolov5m, yolov5x, custom

# Images
img = 'https://ultralytics.com/images/zidane.jpg'  # or file, PIL, OpenCV, numpy, multiple

# Inference
results = model(img, augment=True)  # <--- TTA inference

# Results
results.print()  # or .show(), .save(), .crop(), .pandas(), etc.

Customize

You can customize the TTA ops applied in the YOLOv5 forward_augment() method here:

yolov5/models/yolo.py

Lines 125 to 137 in 8c6f9e1

	def forward_augment(self, x):
	img_size = x.shape[-2:] # height, width
	s = [1, 0.83, 0.67] # scales
	f = [None, 3, None] # flips (2-ud, 3-lr)
	y = [] # outputs
	for si, fi in zip(s, f):
	xi = scale_img(x.flip(fi) if fi else x, si, gs=int(self.stride.max()))
	yi = self.forward_once(xi)[0] # forward
	# cv2.imwrite(f'img_{si}.jpg', 255 * xi[0].cpu().numpy().transpose((1, 2, 0))[:, :, ::-1]) # save
	yi = self._descale_pred(yi, fi, si, img_size)
	y.append(yi)
	return torch.cat(y, 1), None # augmented inference, train

Environments

YOLOv5 may be run in any of the following up-to-date verified environments (with all dependencies including CUDA/CUDNN, Python and PyTorch preinstalled):

• Notebooks with free GPU:
• Google Cloud Deep Learning VM. See GCP Quickstart Guide
• Amazon Deep Learning AMI. See AWS Quickstart Guide
• Docker Image. See Docker Quickstart Guide

Status

If this badge is green, all YOLOv5 GitHub Actions Continuous Integration (CI) tests are currently passing. CI tests verify correct operation of YOLOv5 training, validation, inference, export and benchmarks on MacOS, Windows, and Ubuntu every 24 hours and on every commit.

[swapnil-saha]

I have a question about the value under column P. Is it map@.60 ? (the default IOU threshold value is .60 at test.py)

[Aktcob]

I have a question about the value under column P. Is it map@.60 ? (the default IOU threshold value is .60 at test.py)

The default IOU threshold value is NMS threshold, not the map

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[gizemtanriver]

Hi, what are those 3 different resolutions that TTA uses? Are they selected randomly? Many thanks

[vaskers5]

#10312 pls check my pr - I have fixed this problem

Now it can works like this:

[Robotatron]

does it work with segmentation?