wip: Thu 20 Mar 2025 23:50:48 PDT

research/roi.py

@@ -5,6 +5,7 @@ import os
 import requests
 import json
 import subprocess
+import cv2
 
 class RoiResearch(object):
     def __init__(self, token=None):
@@ -37,13 +38,6 @@ class RoiResearch(object):
                     return
                 yield sd
 
-def parse_args():
-    parser = argparse.ArgumentParser()
-    parser.add_argument("--token", '-t', default='3ebd8c33-f46e-4b06-bda8-4c0f5f5eb530', type=str)
-    parser.add_argument("--username", "-u", type=str)
-    parser.add_argument("--password", "-p", type=str)
-    return parser.parse_args()
-
 def get_last_id(list_file):
     ret = None
     if not os.path.exists(list_file):
@@ -63,9 +57,8 @@ def get_all_ids(list_file):
         for line in f:
             ret.append(json.loads(line)['id'])
     return ret
-    
+
-def main():
+def get_roi_data(args):
-    args = parse_args()
     rr = RoiResearch(args.token)
     if args.username and args.password:
         print(rr.login(args.username, args.password))
@@ -93,6 +86,120 @@ def main():
         cmd = f'../scripts/emcli get-scan-data {id} -o {outd}'
         print(cmd)
         subprocess.call(cmd, shell=True)
+    
+def preprocess(img):
+    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
+    min_val, max_val, min_loc, max_loc = cv2.minMaxLoc(gray)
+    histo = cv2.calcHist([gray], [0], None, [256], [0, 256])
+    total = gray.shape[0] * gray.shape[1]
+    s = 0
+    low_thres = -1
+    high_thres = -1
+    low_proportion = 70
+    high_proportion = 80
+    for x in range(256):
+        i = histo[x]
+        s += i
+        ratio = s / total * 100
+        if low_thres == -1 and ratio > low_proportion:
+            low_thres = x
+        if high_thres == -1 and ratio > high_proportion:
+            high_thres = x
+    if low_thres == -1:
+        raise Exception("no threshold found")
+    print(low_thres, high_thres)
+    for i in range(gray.shape[0]):
+        for j in range(gray.shape[1]):
+            if gray[i, j] < low_thres:
+                gray[i, j] = 0
+            elif gray[i, j] > high_thres:
+                gray[i, j] = 0
+            else:
+                gray[i, j] = 255
+    return gray
+
+window_y = 0
+def show_img(img, name):
+    global window_y
+    cv2.imshow(name, img)
+    cv2.moveWindow(name, 640, window_y)
+    window_y += 200
+
+def find_point(roi_processed, i, j):
+    tolerrance = 2
+    for di in range(-tolerrance, tolerrance + 1):
+        for dj in range(-tolerrance, tolerrance + 1):
+            x = i + di
+            y = j + dj
+            if x < 0 or x >= roi_processed.shape[0] or y < 0 or y >= roi_processed.shape[1]:
+                continue
+            if roi_processed[x, y] > 0:
+                return True
+
+def calc_sim(frame_processed, roi_processed):
+    s = 0
+    total = 0
+    width = frame_processed.shape[1]
+    height = frame_processed.shape[0]
+    for i in range(height):
+        for j in range(width):
+            roi_val = roi_processed[i, j]
+            frame_val = frame_processed[i, j]
+            if roi_val:
+                total += 1
+                if find_point(frame_processed, i, j):
+                    s += 1
+            if frame_val:
+                total += 1
+                if find_point(roi_processed, i, j):
+                    s += 1
+    return s / total
+
+def roi_sim(frame_roi_img, roi_img):
+    if frame_roi_img.shape != roi_img.shape:
+        raise Exception("size mismatch")
+    frame_processed = preprocess(frame_roi_img)
+    roi_processed = preprocess(roi_img)
+    siml = calc_sim(frame_processed, roi_processed)
+    if False:
+        show_img(frame_processed, "frame_processed")
+        show_img(roi_processed, "roi_processed")
+        show_img(frame_roi_img, "frame_roi_img")
+        show_img(roi_img, "roi_img")
+        cv2.waitKey(0)
+        cv2.destroyAllWindows()
+    return siml
+
+def process_roi_data(id):
+    rd = f"data/roi/samples/{id}"
+    frame_file = os.path.abspath(f"{rd}/{id}-frame.jpg")
+    roi_file = os.path.abspath(f"{rd}/{id}-roi.jpg")
+    roi_img = cv2.imread(roi_file)
+    cmd = f'./qrtool frame_roi {frame_file}'
+    subprocess.check_call(cmd, shell=True, cwd=os.path.dirname(os.path.abspath(__file__)) + "/../alg")
+    frame_roi_img = cv2.imread(frame_file + ".roi.jpg")
+    size = [128, 128]
+    frame_roi_img = cv2.resize(frame_roi_img, size)
+    roi_img = cv2.resize(roi_img, size)
+    sim = roi_sim(frame_roi_img, roi_img)
+    print(f"{id}: {sim}")
+
+def parse_args():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--token", '-t', default='3ebd8c33-f46e-4b06-bda8-4c0f5f5eb530', type=str)
+    parser.add_argument("--username", "-u", type=str)
+    parser.add_argument("--password", "-p", type=str)
+    parser.add_argument("--download", "-d", action='store_true')
+    parser.add_argument("--id", "-i", type=int, action='append')
+    return parser.parse_args()
+
+def main():
+    args = parse_args()
+    if args.download:
+        get_roi_data(args)
+    if args.id:
+        for id in args.id:
+            process_roi_data(id)
 
 if __name__ == "__main__":
     main()