Update evaluate.py

evaluate.py

@@ -70,14 +70,13 @@ def load_json(path):
 def eval(test, standard, tmpdir):
     test_p = tmpdir + "/user.mp4"
     standard_p = tmpdir + "/standard.mp4"
-    os.system('python inferencer_demo.py ' + test_p + ' --pred-out-dir', tmpdir) # produce user.json
+    os.system('python inferencer_demo.py ' + test_p + ' --pred-out-dir ' + tmpdir) # produce user.json
 
     scores = []
 
     align_filter(tmpdir + '/standard', tmpdir + '/user', tmpdir) # 帧对齐 produce aligned vedios
 
-    # return None
-    data_00 = load_json(tmpdir + '/standard.json') # aligned json
+    data_00 = load_json(tmpdir + '/standard.json') 
     data_01 = load_json(tmpdir + '/user.json')
     cap_00 = cv2.VideoCapture(standard_p)
     cap_01 = cv2.VideoCapture(test_p)
@@ -160,7 +159,6 @@ def eval(test, standard, tmpdir):
                     start_point = (int(keypoints_01[start][0]), int(keypoints_01[start][1]))
                     end_point = (int(keypoints_01[end][0]), int(keypoints_01[end][1]))
                     cur_score = findcos_single([[int(keypoints_01[start][0]), int(keypoints_01[start][1])], [int(keypoints_01[end][0]), int(keypoints_01[end][1])]], [[int(keypoints_00_ori[start][0]), int(keypoints_00_ori[start][1])], [int(keypoints_00_ori[end][0]), int(keypoints_00_ori[end][1])]])
-                    # print(cur_score[0])
 
                      # 如果当前相似度小于 99.3，认为有误差，并记录下来
                     if float(cur_score[0]) < 98.8 and start != 5:
@@ -171,7 +169,6 @@ def eval(test, standard, tmpdir):
                             bigerror.append(start)
                     else:
                         cv2.line(frame_01, start_point, end_point, (255, 0, 0), line_width)  # Blue line
-                    # cv2.line(frame_01, start_point, end_point, (255, 0, 0), line_width)  # Blue line
 
             for (start, end) in connections2:
                 start = start - 1
@@ -180,9 +177,9 @@ def eval(test, standard, tmpdir):
                     if i < len(keypoints_01) and i + 1 < len(keypoints_01):
                         start_point = (int(keypoints_01[i][0]), int(keypoints_01[i][1]))
                         end_point = (int(keypoints_01[i + 1][0]), int(keypoints_01[i + 1][1]))
-                        # cv2.line(frame_01, start_point, end_point, (255, 0, 0), line_width)  # Blue line
+
                         cur_score = findcos_single([[int(keypoints_01[i][0]), int(keypoints_01[i][1])], [int(keypoints_01[i + 1][0]), int(keypoints_01[i + 1][1])]], [[int(keypoints_00_ori[i][0]), int(keypoints_00_ori[i][1])], [int(keypoints_00_ori[i + 1][0]), int(keypoints_00_ori[i + 1][1])]])
-                        # print(cur_score[0])
+
                         if float(cur_score[0]) < 98.8:
                             error.append(start)
                             cv2.line(frame_01, start_point, end_point, (0, 0, 255), 2)  # Red line
@@ -202,23 +199,12 @@ def eval(test, standard, tmpdir):
 
         if frame_id_00 < min_length and frame_id_01 < min_length:
             min_cos, min_idx = findCosineSimilarity_1(data_00[frame_id_00]["instances"][0]["keypoints"], data_01[frame_id_01]["instances"][0]["keypoints"])
-            # print(min_cos)
-            # if min_cos < 99:
-            #     cv2.putText(combined_frame, "Incorrect Gesture", (120, 220),
-            #                 cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 0, 255), 2)
-            # else:
-            #     cv2.putText(combined_frame, "Correct Gesture", (120, 220),
-            #                 cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 0), 2)
 
          # 如果存在误差，将误差部分对应的人体部位加入内容列表
         if error != []:
             # print(error)
             content = []
             for i in error:
-                # if i in [5,7]: content.append('Right Arm')
-                # if i in [6,8]: content.append('Left Arm')
-                # if i > 90 and i < 112: content.append('Right Hand')
-                # if i >= 112: content.append('Left Hand')
                 if i in [5,7]: content.append('Left Arm')
                 if i in [6,8]: content.append('Right Arm')
                 if i > 90 and i < 112: content.append('Left Hand')
@@ -226,7 +212,6 @@ def eval(test, standard, tmpdir):
             part = ""
 
             # 在视频帧上显示检测到的误差部位
-            # cv2.putText(combined_frame, "Please check: ", (430, 30), cv2.FONT_HERSHEY_SIMPLEX, 0.3, (0, 0, 255), line_width)
             cv2.putText(combined_frame, "Please check: ", (int(frame_width*1.75), int(frame_height*0.2)), cv2.FONT_HERSHEY_SIMPLEX, 1.2, (0, 0, 255), 2)
             start_x = int(frame_width*1.75) + 10   #435 # 起始的 x 坐标
             start_y = int(frame_height*0.2) + 50 # 45
@@ -242,10 +227,6 @@ def eval(test, standard, tmpdir):
         if bigerror != []:
             bigcontent = []
             for i in bigerror:
-                # if i in [5,7]: bigcontent.append('Right Arm')
-                # if i in [6,8]: bigcontent.append('Left Arm')
-                # if i > 90 and i < 112: bigcontent.append('Right Hand')
-                # if i >= 112: bigcontent.append('Left Hand')
                 if i in [5,7]: bigcontent.append('Left Arm')
                 if i in [6,8]: bigcontent.append('Right Arm')
                 if i > 90 and i < 112: bigcontent.append('Left Hand')
@@ -257,10 +238,8 @@ def eval(test, standard, tmpdir):
         cnt += 1
         combined_frame = np.vstack((combined_frame_ori, combined_frame))
         out.write(combined_frame)
-        # print(f"min_cos: {float(min_cos)}")
         scores.append(float(min_cos)) # 记录每一帧的相似度得分
 
-    # print(f"scores: {scores}")
     fps = 5  # Frames per second
     frame_numbers = list(error_dict.keys())  # List of frame numbers 获取含有严重误差的帧号列表
     time_intervals = [(frame / fps, (frame + 1) / fps) for frame in frame_numbers] # 将帧号转换为时间区间（秒）
@@ -272,4 +251,4 @@ def eval(test, standard, tmpdir):
     # 1. scores 的平均值，作为整体手势相似度的评分
     # 2. final_merged_intervals，合并后的误差时间区间及其对应的误差信息
     # 3. comments，用于给用户的速度建议（加快或放慢手势）
-    return sum(scores) / len(scores), final_merged_intervals, comments
+    return sum(scores) / len(scores), final_merged_intervals, comments