/PaTTA

A test times augmentation toolkit based on paddle2.0.

Primary LanguagePythonMIT LicenseMIT

Patta

GitHub forks GitHub Repo stars GitHub Upload Python Package PaTTA Tests

Image Test Time Augmentation with Paddle2.0!

           Input
             |           # input batch of images 
        / / /|\ \ \      # apply augmentations (flips, rotation, scale, etc.)
       | | | | | | |     # pass augmented batches through model
       | | | | | | |     # reverse transformations for each batch of masks/labels
        \ \ \|/ / /      # merge predictions (mean, max, gmean, etc.)
             |           # output batch of masks/labels
           Output

Table of Contents

  1. Quick Start
  1. Transforms
  2. Aliases
  3. Merge modes
  4. Installation

Quick start (Default Transforms)

Test

We support that you can use the following to test after defining the network.

Segmentation model wrapping [docstring]:
import patta as tta
tta_model = tta.SegmentationTTAWrapper(model, tta.aliases.d4_transform(), merge_mode='mean')
Classification model wrapping [docstring]:
tta_model = tta.ClassificationTTAWrapper(model, tta.aliases.five_crop_transform())
Keypoints model wrapping [docstring]:
tta_model = tta.KeypointsTTAWrapper(model, tta.aliases.flip_transform(), scaled=True)

Note: the model must return keypoints in the format Tensor([x1, y1, ..., xn, yn])

Predict

We support that you can use the following to test when you have the static model: *.pdmodel、*.pdiparams、*.pdiparams.info.

Load model [docstring]:
import patta as tta
model = tta.load_model(path='output/model')
Segmentation model wrapping [docstring]:
tta_model = tta.SegmentationTTAWrapper(model, tta.aliases.d4_transform(), merge_mode='mean')
Classification model wrapping [docstring]:
tta_model = tta.ClassificationTTAWrapper(model, tta.aliases.five_crop_transform())
Keypoints model wrapping [docstring]:
tta_model = tta.KeypointsTTAWrapper(model, tta.aliases.flip_transform(), scaled=True)

Use-Tools

Segmentation model [docstring]:

We recommend modifying the file seg.py according to your own model.

python seg.py --model_path='output/model' \
                 --batch_size=16 \
                 --test_dataset='test.txt'

Note: Related to paddleseg

Advanced-Examples (DIY Transforms)

Custom transform:
# defined 2 * 2 * 3 * 3 = 36 augmentations !
transforms = tta.Compose(
    [
        tta.HorizontalFlip(),
        tta.Rotate90(angles=[0, 180]),
        tta.Scale(scales=[1, 2, 4]),
        tta.Multiply(factors=[0.9, 1, 1.1]),        
    ]
)

tta_model = tta.SegmentationTTAWrapper(model, transforms)
Custom model (multi-input / multi-output)
# Example how to process ONE batch on images with TTA
# Here `image`/`mask` are 4D tensors (B, C, H, W), `label` is 2D tensor (B, N)

for transformer in transforms: # custom transforms or e.g. tta.aliases.d4_transform() 
    
    # augment image
    augmented_image = transformer.augment_image(image)
    
    # pass to model
    model_output = model(augmented_image, another_input_data)
    
    # reverse augmentation for mask and label
    deaug_mask = transformer.deaugment_mask(model_output['mask'])
    deaug_label = transformer.deaugment_label(model_output['label'])
    
    # save results
    labels.append(deaug_mask)
    masks.append(deaug_label)
    
# reduce results as you want, e.g mean/max/min
label = mean(labels)
mask = mean(masks)

Optional Transforms

Transform Parameters Values
HorizontalFlip - -
VerticalFlip - -
HorizontalShift shifts List[float]
VerticalShift shifts List[float]
Rotate90 angles List[0, 90, 180, 270]
Scale scales
interpolation		 List[float]
"nearest"/"linear"
Resize sizes
original_size
interpolation		 List[Tuple[int, int]]
Tuple[int,int]
"nearest"/"linear"
Add values List[float]
Multiply factors List[float]
FiveCrops crop_height
crop_width		 int
int
AdjustContrast factors List[float]
AdjustBrightness factors List[float]
AverageBlur kernel_sizes List[Union[Tuple[int, int], int]]
GaussianBlur kernel_sizes
sigma		 List[Union[Tuple[int, int], int]]
Optional[Union[Tuple[float, float], float]]
Sharpen kernel_sizes List[int]

Aliases (Combos)

  • flip_transform (horizontal + vertical flips)
  • hflip_transform (horizontal flip)
  • d4_transform (flips + rotation 0, 90, 180, 270)
  • multiscale_transform (scale transform, take scales as input parameter)
  • five_crop_transform (corner crops + center crop)
  • ten_crop_transform (five crops + five crops on horizontal flip)

Merge-modes

Installation

PyPI:

# Use pip install PaTTA
$ pip install patta

or

# After downloading the whole dir
$ git clone https://github.com/AgentMaker/PaTTA.git
$ pip install PaTTA/

Run tests

python -m pytest

Contact us

Email : agentmaker@163.com
QQ Group : 1005109853