first commit

README

@@ -0,0 +1,10 @@
+Simple image processor.
+
+Will read from ./frames and output in ./output
+
+Run 'bash process.sh' for bulk conversion, with 3000 circles and seed of 1 on 8 threads.
+
+![screenshot](output/image1.jpg.png)
+![screenshot](output/image1_pixel.jpg.png)
+![screenshot](output/image2.png.png)
+![screenshot](output/image2_pixel.png.png)

process.sh

@@ -0,0 +1,18 @@
+#!/bin/bash
+
+# Number of parallel jobs
+N=8
+
+# Loop through the images in chunks of $N
+for ((i=0; i<$(ls frames/ | wc -l); i+=N)); do
+  # Launch $N jobs in parallel
+  for ((j=0; j<N; j++)); do
+    idx=$((i+j))
+    if [ $idx -lt $(ls frames/ | wc -l) ]; then
+      python processor.py "frames/$(ls frames/ | sed -n $((idx+1))p)" "output/$(ls frames/ | sed -n $((idx+1))p)" 3000 1&
+    fi
+  done
+
+  # Wait for the $N jobs to finish before launching the next batch
+  wait
+done

processor.py

@@ -0,0 +1,69 @@
+import numpy as np
+import cv2
+import argparse
+import time
+from numpy import random
+
+def load_image(image_path):
+    image = cv2.imread(image_path, cv2.IMREAD_COLOR)
+    if image is None:
+        raise ValueError(f"Image at path {image_path} could not be loaded.")
+    return image.astype(np.float32)
+
+def draw_circle(image, reference_image, center, radius):
+    mask = np.zeros(image.shape[:2], dtype=np.uint8)
+    cv2.circle(mask, center, int(radius), 255, -1)
+
+    mean_color = cv2.mean(reference_image, mask=mask)[:3]
+    mean_color = tuple(map(int, mean_color))
+
+    tmp_image = image.copy()
+    cv2.circle(tmp_image, center, int(radius), mean_color, -1)
+    return tmp_image
+
+def calculate_vector_distance(image1, image2):
+    return np.linalg.norm(image1 - image2)
+
+def main(in_path, out_path, iterations, seed):
+    random.seed(seed)
+    image1 = load_image(in_path)
+    height, width, _ = image1.shape
+    image2 = np.zeros((height,width,3), np.uint8)
+
+    last_time = time.time()
+
+    for i in range(iterations):
+        print(f"Image {in_path} pass {i}", end="\r")
+
+        center = (random.randint(width), random.randint(height))
+        radius = max(width, height) / 2
+        step = radius
+
+        for _ in range(int(np.log2(max(width, height))) + 1):
+            candidate_up = draw_circle(image2, image1, center, radius + step)
+            candidate_up_distance = calculate_vector_distance(image1, candidate_up)
+
+            candidate_down = draw_circle(image2, image1, center, max(0, radius - step))
+            candidate_down_distance = calculate_vector_distance(image1, candidate_down)
+
+            if candidate_down_distance <= candidate_up_distance:
+                radius = max(0, radius - step)
+            else:
+                radius += step
+
+            step /= 2
+
+        image2 = draw_circle(image2, image1, center, radius)
+
+    print(f"{out_path} took {-last_time + time.time()} seconds")
+    cv2.imwrite(f"{out_path}.png", np.uint8(image2))
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description="Draw and optimize a circle on an image using gradient ascent.")
+    parser.add_argument("input", help="Path to the reference image.")
+    parser.add_argument("output", help="Path to save the output image.")
+    parser.add_argument("circles", help="The amount of circles to draw.")
+    parser.add_argument("seed", help="The seed to use.")
+
+    args = parser.parse_args()
+    main(args.input, args.output, int(args.circles), int(args.seed))