Autolabel script, Research notebook, COCO annotations for raw samples

autolabel.py

@@ -0,0 +1,206 @@
+import os
+import sys
+from math import sqrt
+from skimage.io import imread
+from skimage.feature import blob_dog, blob_log, blob_doh
+from skimage.color import rgb2gray
+import base64
+import pathlib
+import json
+from PIL import Image
+from argparse import ArgumentParser
+import matplotlib.pyplot as plt
+
+
+def main(target_dir_or_file, max_sigma=5, thresh=0.1, pick_index=None, show=True, save=False):
+    '''This script uses blob detection features of skimage (with parameters set by the user) to
+    automatically generate annotations for quantum dots in the COCO format. It can operate on
+    single image files, or recursively operate on directories of images.'''
+    if os.path.isdir(target_dir_or_file):
+        process_image_dir(target_dir_or_file, max_sigma, thresh, pick_index, show, save)
+    elif os.path.isfile(target_dir_or_file):
+        process_single_file(target_dir_or_file, max_sigma, thresh, pick_index, show, save)
+    else:
+        print("invalid image_dir or filepath, please check that this path exists or use the --help flag for usage information.")
+
+def to_single_coco_label(x,y,r):
+    '''
+    Takes in circular label information and transforms it into a COCO label in polygon format with a bounding box.
+
+    :int x: x-coordinate of circle's center
+    :int y: y-coordinate of circle's center
+    :int r: radius of circle
+
+    Returns dict object containing a single label in COCO format.
+    '''
+    x1, y1 = x-r, y+r # top left
+    x2, y2 = x+r, y+r # top right
+    x3, y3 = x-r, y-r # bottom left
+    x4, y4 = x+r, y-r # bottom right
+    return {
+      "segmentation": [[x1,y1,x2,y2,x3,y3,x4,y4]], # list of points in polygon (just a rectangle in this case)
+      "bbox": [x1,y1,x4,y4], # bounding box in 
+      "group_id": "null",
+      "shape_type": "polygon",
+      "flags": {}
+    }
+
+def to_coco_annotations_file(filepath, labels):
+    '''
+    Takes in a single filepath and a list of all labels for it, and
+    generates a full COCO annotations file for it.
+
+    :str filepath: filepath to raw image
+    :list labels: list of labels (each label generated by the `to_single_coco_label()` function)
+    '''
+    with open(filepath,'rb') as f:
+        img_data = base64.b64encode(f.read()).decode('utf-8')
+    image = Image.open(filepath)
+    img_width, img_height = image.size
+
+    return {
+        "version": "4.6.0",
+        "flags": {},
+        "annotations": labels,
+        "imagePath": filepath,
+        "imageData": img_data,
+        "imageWidth": img_width,
+        "imageHeight": img_height
+    }
+
+def process_image_dir(image_dir: str, max_sigma: float, thresh: float, pick_index: int, show: bool, save: bool):
+    for curr_dir, next_dirs, files in os.walk(image_dir):
+        for f in files:
+            if not f.endswith('.tif'):
+                continue
+
+            filepath = os.path.join(curr_dir, f)
+            image = image_gray = imread(filepath, as_gray=True)
+
+            blobs_log = blob_log(image_gray, max_sigma=max_sigma, num_sigma=10, threshold=thresh)
+
+            # Compute radii in the 3rd column.
+            blobs_log[:, 2] = blobs_log[:, 2] * sqrt(2)
+
+            blobs_dog = blob_dog(image_gray, max_sigma=max_sigma, threshold=thresh)
+            blobs_dog[:, 2] = blobs_dog[:, 2] * sqrt(2)
+
+            blobs_doh = blob_doh(image_gray, max_sigma=max_sigma, threshold=thresh)
+
+            blobs_list = [blobs_log, blobs_dog, blobs_doh]
+            colors = ['yellow', 'lime', 'red']
+            titles = ['Laplacian of Gaussian', 'Difference of Gaussian',
+                    'Determinant of Hessian']
+            sequence = zip(blobs_list, colors, titles)
+
+            fig, axes = plt.subplots(1, 3, figsize=(9, 3), sharex=True, sharey=True)
+            ax = axes.ravel()
+            labels = []
+            for idx, (blobs, color, title) in enumerate(sequence):
+                ax[idx].set_title(title)
+                ax[idx].imshow(image)
+                for blob in blobs:
+                    y, x, r = blob
+                    c = plt.Circle((x, y), r, color=color, linewidth=2, fill=False)
+
+                    if idx == pick_index:
+                        label = to_single_coco_label(x, y, r)
+                        labels.append(label)
+
+                    ax[idx].add_patch(c)
+                ax[idx].set_axis_off()
+
+            if show:
+                plt.tight_layout()
+                plt.show()
+                sys.exit(0)
+            elif save:
+                out = to_coco_annotations_file(filepath, labels)
+                p = pathlib.Path(filepath)
+                output_path = os.path.join(curr_dir, p.stem + '.json')
+                with open(output_path, 'w') as f:
+                    json.dump(out, f)
+                print(f"Saved to {output_path}")
+
+def process_single_file(filepath: str, max_sigma: float, thresh: float, pick_index: int, show: bool, save: bool):
+    image = image_gray = imread(filepath, as_gray=True)
+
+    blobs_log = blob_log(image_gray, max_sigma=max_sigma, num_sigma=10, threshold=thresh)
+
+    # Compute radii in the 3rd column.
+    blobs_log[:, 2] = blobs_log[:, 2] * sqrt(2)
+
+    blobs_dog = blob_dog(image_gray, max_sigma=max_sigma, threshold=thresh)
+    blobs_dog[:, 2] = blobs_dog[:, 2] * sqrt(2)
+
+    blobs_doh = blob_doh(image_gray, max_sigma=max_sigma, threshold=thresh)
+
+    blobs_list = [blobs_log, blobs_dog, blobs_doh]
+    colors = ['yellow', 'lime', 'red']
+    titles = ['Laplacian of Gaussian', 'Difference of Gaussian',
+            'Determinant of Hessian']
+    sequence = zip(blobs_list, colors, titles)
+
+    fig, axes = plt.subplots(1, 3, figsize=(9, 3), sharex=True, sharey=True)
+    ax = axes.ravel()
+    labels = []
+    for idx, (blobs, color, title) in enumerate(sequence):
+        ax[idx].set_title(title)
+        ax[idx].imshow(image)
+        for blob in blobs:
+            y, x, r = blob
+            c = plt.Circle((x, y), r, color=color, linewidth=2, fill=False)
+
+            if idx == pick_index:
+                label = to_single_coco_label(x, y, r)
+                labels.append(label)
+
+            ax[idx].add_patch(c)
+        ax[idx].set_axis_off()
+
+    # show preview of algos with given parameters on example image
+    if show:
+        plt.tight_layout()
+        plt.show()
+        sys.exit(0)
+
+    # save 
+    elif save:
+        out = to_coco_annotations_file(filepath, labels)
+        p = pathlib.Path(filepath)
+        curr_dir = os.path.dirname(filepath)
+        output_path = os.path.join(curr_dir, p.stem + '.json')
+        with open(output_path, 'w') as f:
+            json.dump(out, f)
+        print(f"Saved to {output_path}")
+
+if __name__ == '__main__':
+    parser = ArgumentParser("autolabel-parser", description="""This script uses blob detection features of skimage (with parameters set by the user) to
+    automatically generate annotations for quantum dots in the COCO format. It can operate on
+    single image files, or recursively operate on directories of images.""")
+    parser.add_argument('--target', type=str, required=True, help='image directory or filename')
+    parser.add_argument('--max_sigma', type=int, default=5, required=False, help='The maximum standard deviation for Gaussian kernel. Keep this high to detect larger blobs. Range of 1-20 recommended.')
+    parser.add_argument('--threshold', type=float, default=0.1, required=False, help='The absolute lower bound for scale space maxima. Local maxima smaller than threshold are ignored. Reduce this to detect blobs with lower intensities. Range 0.0-0.2 recommended.')
+    parser.add_argument('--show', action="store_true", help="flag to show preview of all 3 algorithms with the given settings on an example image.")
+    parser.add_argument('--save', action="store_true", help="flag to use the specified algorithm (LoG, DoG, DoH) to label all images and save labels as JSON files.")
+    parser.add_argument('--log', action="store_true", help="Laplacian of Gaussian (leftmost image in --show mode")
+    parser.add_argument('--dog', action="store_true", help="Difference of Gaussian (middle image in --show mode)")
+    parser.add_argument('--doh', action="store_true", help="Determinant of Hessian (rightmost image in --show mode)")
+    args = parser.parse_args()
+    print(f"Autolabel parameters: {args}")
+
+    # validate algo if --save specified
+    pick_index = None
+    if args.save:
+        if args.log:
+            pick_index = 0
+        elif args.dog:
+            pick_index = 1
+        elif args.doh:
+            pick_index = 2
+        else:
+            print("You must choose an algorithm (--log, --dog, or --doh) to save annotations. Use --help for more details.")
+            sys.exit(0)
+
+    # execute script with these params
+    main(args.target, args.max_sigma, args.threshold, pick_index, args.show, args.save)

find_dots4_history.txt

@@ -0,0 +1,20 @@
+python3 find_dots4.py input_dir/32 5 0.1 0 save
+python3 find_dots4.py input_dir/106 5 0.2 1 save
+python3 find_dots4.py input_dir/106/2a.tif 5 0.1 1 save
+python3 find_dots4.py input_dir/106/2b.tif 5 0.1 1 save
+python3 find_dots4.py input_dir/106/2c.tif 5 0.1 1 save
+python3 find_dots4.py input_dir/106/5a.tif 3 0.05 0 save
+python3 find_dots4.py input_dir/107 5 0.1 1 save
+python3 find_dots4.py input_dir/107/5a.tif 5 0.1 0 save
+python3 find_dots4.py input_dir/132 5 0.1 0 save
+python3 find_dots4.py input_dir/157 5 0.1 1 save
+python3 find_dots4.py input_dir/157/5a.tif 5 0.05 1 save
+python3 find_dots4.py input_dir/404 5 0.1 1 save
+python3 find_dots4.py input_dir/405 5 0.1 1 save
+python3 find_dots4.py input_dir/406 10 0.1 1 save
+python3 find_dots4.py input_dir/86 5 0.2 1 save
+python3 find_dots4.py input_dir/86/2n.tif 20 0.1 1 save
+python3 find_dots4.py input_dir/86/2o.tif 20 0.1 1 save
+python3 find_dots4.py input_dir/86/2p.tif 20 0.2 1 save
+python3 find_dots4.py input_dir/87 20 0.15 1 save
+python3 find_dots4.py input_dir/87/5a.tif 20 0.1 0 save

generate_labels.sh

@@ -0,0 +1,21 @@
+#!/bin/bash
+python3 autolabel.py --target="input_dir/32" --max_sigma=5 --threshold=0.1 --log --save &
+python3 autolabel.py --target="input_dir/106" --max_sigma=5 --threshold=0.2 --dog --save &
+python3 autolabel.py --target="input_dir/106/2a.tif" --max_sigma=5 --threshold=0.1 --dog --save &
+python3 autolabel.py --target="input_dir/106/2b.tif" --max_sigma=5 --threshold=0.1 --dog --save &
+python3 autolabel.py --target="input_dir/106/2c.tif" --max_sigma=5 --threshold=0.1 --dog --save &
+python3 autolabel.py --target="input_dir/106/5a.tif" --max_sigma=3 --threshold=0.05 --log --save &
+python3 autolabel.py --target="input_dir/107" --max_sigma=5 --threshold=0.1 --dog --save &
+python3 autolabel.py --target="input_dir/107/5a.tif" --max_sigma=5 --threshold=0.1 --log --save &
+python3 autolabel.py --target="input_dir/132" --max_sigma=5 --threshold=0.1 --log --save &
+python3 autolabel.py --target="input_dir/157" --max_sigma=5 --threshold=0.1 --dog --save &
+python3 autolabel.py --target="input_dir/157/5a.tif" --max_sigma=5 --threshold=0.05 --dog --save &
+python3 autolabel.py --target="input_dir/404" --max_sigma=5 --threshold=0.1 --dog --save &
+python3 autolabel.py --target="input_dir/405" --max_sigma=5 --threshold=0.1 --dog --save &
+python3 autolabel.py --target="input_dir/406" --max_sigma=10 --threshold=0.1 --dog --save &
+python3 autolabel.py --target="input_dir/86" --max_sigma=5 --threshold=0.2 --dog --save &
+python3 autolabel.py --target="input_dir/86/2n.tif" --max_sigma=20 --threshold=0.1 --dog --save &
+python3 autolabel.py --target="input_dir/86/2o.tif" --max_sigma=20 --threshold=0.1 --dog --save &
+python3 autolabel.py --target="input_dir/86/2p.tif" --max_sigma=20 --threshold=0.2 --dog --save &
+python3 autolabel.py --target="input_dir/87" --max_sigma=20 --threshold=0.15 --dog --save &
+python3 autolabel.py --target="input_dir/87/5a.tif" --max_sigma=20 --threshold=0.1 --log --save &