BloodAxe

/**
 * @brief   Mapping functions from Eigen data types to OpenCV
 * @author  Eugene Khvedchenya <ekhvedchenya@gmail.com>
 * @details This header file contains code snippet for easy mapping Eigen types to OpenCV and back with minimal overhead.
 * @more    computer-vision.talks.com/articles/mapping-eigen-to-opencv/
 * Features:
 *  - Mapping plain data types with no overhead (read/write access)
 *  - Mapping expressions via evaluation (read only acess)
 *
 * Known issues:

BloodAxe

class RetinaNet(nn.Module):
  RETINA_NET_OUTPUT_BBOXES = "bboxes"
  RETINA_NET_OUTPUT_SCORES = "scores"

  ...

  def forward(self, image):
    x = self.encoder(image)
    x = self.decoder(x)
    bboxes, scores = self.head(x)

BloodAxe

from collections import OrderedDict
from functools import partial
from typing import Union, List, Dict, Tuple, Type

from pytorch_toolbelt.modules import (
    conv1x1,
    UnetBlock,
    ACT_RELU,
    ABN,
    ACT_SWISH,

BloodAxe

import torch

def main():
    torch.autograd.detect_anomaly()
    ...
    # Rest of the training code
   

# OR
class MyNumericallyUnstableLoss(nn.Module):

BloodAxe

# https://github.com/BloodAxe/Kaggle-2020-Alaska2

callbacks += [
  CriterionCallback(
    input_key=INPUT_TRUE_MODIFICATION_FLAG,
    output_key=OUTPUT_PRED_MODIFICATION_FLAG,
    criterion_key="bce"
  ),
  CriterionCallback(
    input_key=INPUT_TRUE_MODIFICATION_TYPE,

BloodAxe

# https://github.com/BloodAxe/Kaggle-2020-Alaska2/blob/master/alaska2/dataset.py#L373
class TrainingValidationDataset(Dataset):
    def __init__(
        self,
        images: Union[List, np.ndarray],
        targets: Optional[Union[List, np.ndarray]],
        quality: Union[List, np.ndarray],
        bits: Optional[Union[List, np.ndarray]],
        transform: Union[A.Compose, A.BasicTransform],
        features: List[str],

BloodAxe

# Bad practice, don't return tuple
class RetinaNet(nn.Module):
  ...

  def forward(self, image):
    x = self.encoder(image)
    x = self.decoder(x)
    bboxes, scores = self.head(x)
    return bboxes, scores

BloodAxe

# https://github.com/BloodAxe/Kaggle-2020-Alaska2/blob/master/alaska2/models/timm.py#L104

def forward(self, **kwargs):
  x = kwargs[self.input_key]
  x = self.rgb_bn(x)
  x = self.encoder.forward_features(x)
  embedding = self.pool(x)
  result = {
    OUTPUT_PRED_MODIFICATION_FLAG: self.flag_classifier(self.drop(embedding)),
    OUTPUT_PRED_MODIFICATION_TYPE: self.type_classifier(self.drop(embedding)),

BloodAxe

class MySegmentationDataset(Dataset):
  ...
  def __getitem__(self, index):
    image = cv2.imread(self.images[index])
    target = cv2.imread(self.masks[index])
    
    # No data normalization and type casting here
    return torch.from_numpy(image).permute(2,0,1).contiguous(),
           torch.from_numpy(target).permute(2,0,1).contiguous()

BloodAxe

class RAMDataset(Dataset):
  def __init__(image_fnames, targets):
    self.targets = targets
    self.images = []
    for fname in tqdm(image_fnames, desc="Loading files in RAM"):
      with open(fname, "rb") as f:
        self.images.append(f.read())
    
  def __len__(self):
    return len(self.targets)