Merge branch 'wildfusion' of https://github.com/WildlifeDatasets/wild…

…life-tools into wildfusion

.github/workflows/code-quality.yml

@@ -25,7 +25,7 @@ jobs:
           python -m pip install --upgrade pip
           pip install black[jupyter]==22.3.0
       - name: Analysing the code with black
-        run: black --check .
+        run: black --check --diff --line-length 120 wildlife_tools
   isort:
     runs-on: ubuntu-latest
     steps:
@@ -40,7 +40,8 @@ jobs:
           python -m pip install --upgrade pip
           pip install isort==5.10.1
       - name: Analysing the code with isort
-        run: isort --check .
+        run: isort --check --diff --line-length 120 wildlife_tools
+
   # flake8:
   #   runs-on: ubuntu-latest
   #   steps:
@@ -55,4 +56,4 @@ jobs:
   #         python -m pip install --upgrade pip
   #         pip install flake8==4.0.1 flake8-docstrings==1.6.0
   #     - name: Analysing the code with flake8
-  #       run: flake8 .
+  #       run: flake8 wildlife_tools

pyproject.toml

@@ -61,4 +61,11 @@ packages = [
 ]
 
 [tool.flake8]
-max-line-length = 120
+max-line-length = 120
+
+[tool.black]
+line-length = 120
+
+[tool.isort]
+profile = "black"
+line_length = 120

wildlife_tools/data/__init__.py

@@ -1 +1 @@
-from .dataset import FeatureDatabase, FeatureDataset, WildlifeDataset, ImageDataset
+from .dataset import FeatureDatabase, FeatureDataset, ImageDataset, WildlifeDataset

wildlife_tools/data/dataset.py

@@ -2,12 +2,12 @@
 import os
 import pickle
 from typing import Callable
+
 import cv2
 import numpy as np
 import pandas as pd
 import pycocotools.mask as mask_coco
 from PIL import Image
-import torch
 
 
 class ImageDataset:
@@ -44,9 +44,7 @@ def __init__(
         self.col_path = col_path
         self.col_label = col_label
         self.load_label = load_label
-        self.labels, self.labels_map = pd.factorize(
-            self.metadata[self.col_label].values
-        )
+        self.labels, self.labels_map = pd.factorize(self.metadata[self.col_label].values)
 
     @property
     def labels_string(self):
@@ -82,7 +80,6 @@ def __getitem__(self, idx):
             return img
 
 
-
 class WildlifeDataset(ImageDataset):
     """
     PyTorch-style dataset for a datasets from wildlife-datasets library.
@@ -122,9 +119,7 @@ def __init__(
         self.col_path = col_path
         self.col_label = col_label
         self.load_label = load_label
-        self.labels, self.labels_map = pd.factorize(
-            self.metadata[self.col_label].values
-        )
+        self.labels, self.labels_map = pd.factorize(self.metadata[self.col_label].values)
 
     def __getitem__(self, idx):
         data = self.metadata.iloc[idx]
@@ -146,9 +141,11 @@ def __getitem__(self, idx):
                 w, h = img.size
                 rles = mask_coco.frPyObjects([segmentation], h, w)
                 segmentation = mask_coco.merge(rles)
-            if isinstance(segmentation, dict) and (isinstance(segmentation['counts'], list) or isinstance(segmentation['counts'], np.ndarray)):            
+            if isinstance(segmentation, dict) and (
+                isinstance(segmentation["counts"], list) or isinstance(segmentation["counts"], np.ndarray)
+            ):
                 # Convert uncompressed RLE to compressed RLE
-                h, w = segmentation['size']
+                h, w = segmentation["size"]
                 segmentation = mask_coco.frPyObjects(segmentation, h, w)
 
         if self.img_load in ["bbox"]:
@@ -182,7 +179,7 @@ def __getitem__(self, idx):
 
         # Mask background using segmentation mask and crop to bounding box.
         elif self.img_load == "bbox_mask":
-            if (not np.any(pd.isnull(segmentation))):
+            if not np.any(pd.isnull(segmentation)):
                 mask = mask_coco.decode(segmentation).astype("bool")
                 img = Image.fromarray(img * mask[..., np.newaxis])
                 y_nonzero, x_nonzero, _ = np.nonzero(img)
@@ -197,7 +194,7 @@ def __getitem__(self, idx):
 
         # Hide object using segmentation mask and crop to bounding box.
         elif self.img_load == "bbox_hide":
-            if (not np.any(pd.isnull(segmentation))):
+            if not np.any(pd.isnull(segmentation)):
                 mask = mask_coco.decode(segmentation).astype("bool")
                 img = Image.fromarray(img * ~mask[..., np.newaxis])
                 y_nonzero, x_nonzero, _ = np.nonzero(img)
@@ -234,7 +231,6 @@ def __getitem__(self, idx):
             return img
 
 
-
 class FeatureDataset:
     """
     PyTorch-style dataset for a extracted features. Couples features with metadata.
@@ -267,9 +263,7 @@ def __init__(
         self.features = features
         self.metadata = metadata.reset_index(drop=True)
         self.col_label = col_label
-        self.labels, self.labels_map = pd.factorize(
-            self.metadata[self.col_label].values
-        )
+        self.labels, self.labels_map = pd.factorize(self.metadata[self.col_label].values)
 
     @property
     def labels_string(self):

wildlife_tools/data/split.py

@@ -1,6 +1,7 @@
 import numpy as np
 
-class SplitChunk():
+
+class SplitChunk:
     """Returns split as given nth chunk of metadata from equally sized metadata chunks."""
 
     def __init__(self, chunk=1, chunk_total=1):

wildlife_tools/features/__init__.py

@@ -1,3 +1,3 @@
 from .deep import ClipFeatures, DeepFeatures
+from .local import AlikedExtractor, DiskExtractor, SiftExtractor, SuperPointExtractor
 from .memory import DataToMemory
-from .local import SiftExtractor, SuperPointExtractor, AlikedExtractor, DiskExtractor

wildlife_tools/features/deep.py

@@ -1,7 +1,8 @@
 import torch
 from tqdm import tqdm
 from transformers import CLIPModel, CLIPProcessor
-from wildlife_tools.data import ImageDataset, FeatureDataset
+
+from wildlife_tools.data import FeatureDataset, ImageDataset
 
 
 class DeepFeatures:
@@ -16,14 +17,14 @@ def __init__(
         num_workers: int = 1,
         device: str = "cpu",
     ):
-        '''
+        """
         Args:
             model: Pytorch model used for the feature extraction.
             batch_size: Batch size used for the feature extraction.
             num_workers: Number of workers used for data loading.
             device: Select between cuda and cpu devices.
 
-        '''
+        """
 
         self.batch_size = batch_size
         self.num_workers = num_workers
@@ -81,15 +82,15 @@ def __init__(
         num_workers=1,
         device="cpu",
     ):
-        '''
+        """
         Args:
             model: transformer.CLIPModel. Uses VIT-L backbone by default.
             processor: transformer.CLIPProcessor. Uses VIT-L processor by default.
             batch_size: Batch size used for the feature extraction.
             num_workers: Number of workers used for data loading.
             device: Select between cuda and cpu devices.
 
-        '''
+        """
         if model is None:
             model = CLIPModel.from_pretrained("openai/clip-vit-large-patch14").vision_model
 

wildlife_tools/features/gluefactory_fix.py

@@ -14,8 +14,8 @@
 except ImportError:
     pycolmap = None
 
-from gluefactory.models.utils.misc import pad_to_length
 from gluefactory.models.extractors.sift import filter_dog_point, run_opencv_sift
+from gluefactory.models.utils.misc import pad_to_length
 
 
 def extract_single_image_fix(self, image: torch.Tensor):
@@ -33,9 +33,7 @@ def extract_single_image_fix(self, image: torch.Tensor):
             scores = np.abs(scores) * scales
     elif self.conf.backend == "opencv":
         # TODO: Check if opencv keypoints are already in corner convention
-        keypoints, scores, scales, angles, descriptors = run_opencv_sift(
-            self.sift, (image_np * 255.0).astype(np.uint8)
-        )
+        keypoints, scores, scales, angles, descriptors = run_opencv_sift(self.sift, (image_np * 255.0).astype(np.uint8))
 
     if (descriptors is None) or (descriptors.size == 0):
         descriptors = np.empty(shape=(0, 128), dtype=np.float32)
@@ -89,7 +87,5 @@ def extract_single_image_fix(self, image: torch.Tensor):
         pred["oris"] = pad_to_length(pred["oris"], num_points, -1, mode="zeros")
         pred["descriptors"] = pad_to_length(pred["descriptors"], num_points, -2, mode="zeros")
         if pred["keypoint_scores"] is not None:
-            pred["keypoint_scores"] = pad_to_length(
-                pred["keypoint_scores"], num_points, -1, mode="zeros"
-            )
+            pred["keypoint_scores"] = pad_to_length(pred["keypoint_scores"], num_points, -1, mode="zeros")
     return pred

wildlife_tools/features/local.py

@@ -1,12 +1,13 @@
+import types
+
 import torch
-from tqdm import tqdm
 from gluefactory.models import get_model
 from omegaconf import OmegaConf
+from tqdm import tqdm
+
 from wildlife_tools.data import FeatureDataset, ImageDataset
 
-# Fix https://github.com/cvg/glue-factory/pull/50
-import types
-from .gluefactory_fix import extract_single_image_fix
+from .gluefactory_fix import extract_single_image_fix  # https://github.com/cvg/glue-factory/pull/50
 
 
 class GlueFactoryExtractor:
@@ -42,14 +43,13 @@ def __init__(
         self.device = device
         self.num_workers = num_workers
 
-
     def __call__(self, dataset: ImageDataset) -> FeatureDataset:
         """
         Extract clip features from input dataset and return them as a new FeatureDataset.
         Gluefactory extractors requires with 3 channel RBG tensors scaled to [0, 1].
 
         Args:
-            dataset: Extract features from this dataset. 
+            dataset: Extract features from this dataset.
         Returns:
             feature_dataset: A FeatureDataset containing the extracted features
         """

wildlife_tools/features/memory.py

@@ -1,6 +1,6 @@
 from tqdm import tqdm
 
-from wildlife_tools.data import ImageDataset, FeatureDataset
+from wildlife_tools.data import FeatureDataset, ImageDataset
 
 
 class DataToMemory:

wildlife_tools/inference/classifier.py

@@ -1,4 +1,5 @@
 from __future__ import annotations
+
 import numpy as np
 import pandas as pd
 import torch
@@ -21,7 +22,6 @@ def __init__(self, database_labels: np.array, k: int = 1, return_scores=False):
         self.database_labels = database_labels
         self.return_scores = return_scores
 
-
     def __call__(self, similarity):
         """
         Predicts the label for each query based on the k nearest matches in the database.
@@ -76,7 +76,6 @@ def __call__(self, similarity):
             return preds
 
 
-
 class TopkClassifier:
     """
     Predict top k query labels given nearest matches in the database.
@@ -93,7 +92,6 @@ def __init__(self, database_labels: np.array, k: int = 10, return_all: bool = Fa
         self.database_labels = database_labels
         self.return_all = return_all
 
-
     def __call__(self, similarity):
         """
         Predicts the top k labels for each query based on the similarity matrix.
@@ -134,9 +132,9 @@ def __call__(self, similarity):
             data.append(list(zip(*data_row)))
 
         preds, scores, idx = list(zip(*data))
-        preds = np.array(preds)[:, :self.k]
-        scores = np.array(scores)[:, :self.k]
-        idx = np.array(idx)[:, :self.k]
+        preds = np.array(preds)[:, : self.k]
+        scores = np.array(scores)[:, : self.k]
+        idx = np.array(idx)[:, : self.k]
 
         if self.return_all:
             return preds, scores, idx

wildlife_tools/similarity/__init__.py

@@ -1,4 +1,4 @@
 from .cosine import CosineSimilarity
-from .wildfusion import WildFusion, SimilarityPipeline
+from .pairwise.lightglue import MatchLightGlue
 from .pairwise.loftr import MatchLOFTR
-from .pairwise.lightglue import MatchLightGlue
+from .wildfusion import SimilarityPipeline, WildFusion

wildlife_tools/similarity/calibration.py

@@ -1,10 +1,9 @@
+import matplotlib.pyplot as plt
 import numpy as np
 import pandas as pd
-import matplotlib.pyplot as plt
-
-from sklearn.linear_model import LogisticRegression
-from sklearn.isotonic import IsotonicRegression
 from scipy.interpolate import PchipInterpolator
+from sklearn.isotonic import IsotonicRegression
+from sklearn.linear_model import LogisticRegression
 
 
 class LogisticCalibration:

wildlife_tools/similarity/cosine.py

@@ -2,7 +2,7 @@
 import torch
 import torch.nn.functional as F
 
-from wildlife_tools.data import FeatureDataset
+from ..data import FeatureDataset
 
 
 def cosine_similarity(a, b):

wildlife_tools/similarity/pairwise/base.py

@@ -1,11 +1,13 @@
-from .collectors import CollectCounts
 import itertools
-import torch
-import numpy as np
-import matplotlib.pyplot as plt
+
 import cv2
+import matplotlib.pyplot as plt
+import numpy as np
+import torch
 from tqdm import tqdm
+
 from ...data import FeatureDataset
+from .collectors import CollectCounts
 
 
 def visualise_matches(img0, keypoints0, img1, keypoints1):
@@ -109,7 +111,7 @@ def __init__(
             batch_size: Number of pairs processed in one batch.
             num_workers: Number of workers used for data loading.
             tqdm_silent: If True, progress bar is disabled.
-            collector: Collector object used for storing results. 
+            collector: Collector object used for storing results.
                 By default, CollectCounts(thresholds=[0.5]) is used.
         """
 
@@ -122,7 +124,10 @@ def __init__(
         self.tqdm_kwargs = {"mininterval": 1, "ncols": 100, "disable": tqdm_silent}
 
     def __call__(
-        self, dataset0: FeatureDataset, dataset1: FeatureDataset, pairs: np.array | None = None
+        self,
+        dataset0: FeatureDataset,
+        dataset1: FeatureDataset,
+        pairs: np.array | None = None,
     ):
         """
         Match pairs of features from two feature datasets.