add yolo-world APP and optimize apps' ret button size

Author: Neutree

projects/app_benchmark/benchmarks/benchmark_npu.py

@@ -73,13 +73,22 @@ def benchmark_forward_img(self, items):
         test_items = []
         path_11n = "/root/models/yolo11n.mud"
         path_11s = "/root/models/yolo11s.mud"
+        path_11n_640 = "/root/models/yolo11n_640.mud"
+        path_11s_640 = "/root/models/yolo11s_640.mud"
+        path_11l_640 = "/root/models/yolo11l_640.mud"
         path_5s = "/root/models/yolov5s.mud"
         path_mbn = "/root/models/mobilenetv2.mud"
         detect_img = "/maixapp/share/picture/2024.1.1/test_coco_640.jpg"
         clssify_img = "/maixapp/share/picture/2024.1.1/cat.jpg"
         test_items.append((path_11n, detect_img))
         if os.path.exists(path_11s):
             test_items.append((path_11s, detect_img))
+        if os.path.exists(path_11n_640):
+            test_items.append((path_11n_640, detect_img))
+        if os.path.exists(path_11s_640):
+            test_items.append((path_11s_640, detect_img))
+        if os.path.exists(path_11l_640):
+            test_items.append((path_11l_640, detect_img))
         test_items.append((path_5s, detect_img))
         test_items.append((path_mbn, clssify_img))
 

projects/app_benchmark/benchmarks/benchmark_stress.py

@@ -105,8 +105,10 @@ def run(self):
         def update_res_vars():
             if self.cam.height() > 480:
                 self.font_scale = 2.2
+                self.font_thickness = 2
             else:
                 self.font_scale = 1.4
+                self.font_thickness = 1
             self.font_size = image.string_size("aA!~?/", scale = self.font_scale)
             self.ai_isp_on = app.get_sys_config_kv("npu", "ai_isp", "0")
             self.font_margin = self.font_size.height() // 2
@@ -123,7 +125,6 @@ def update_res_vars():
             back_rect = [0, 0, self.img_back.width() + self.button_padding_2x, self.img_back.height() + self.button_padding_2x]
             self.back_rect_disp = image.resize_map_pos(self.cam.width(), self.cam.height(), self.disp.width(), self.disp.height(), image.Fit.FIT_CONTAIN, back_rect[0], back_rect[1], back_rect[2], back_rect[3])
             self.string_cam_res_rect_disp = image.resize_map_pos(self.cam.width(), self.cam.height(), self.disp.width(), self.disp.height(), image.Fit.FIT_CONTAIN, self.string_cam_res_rect[0], self.string_cam_res_rect[1], self.string_cam_res_rect[2], self.string_cam_res_rect[3])
-            print(back_rect, self.back_rect_disp, self.cam.width(), self.cam.height(), self.disp.width(), self.disp.height())
         update_res_vars()
         touch_pressed = False
         cam_res_curr_idx = 0
@@ -187,6 +188,9 @@ def update_res_vars():
                 last = (x, y)
                 last_fps = (x, y2)
                 x += padding
+            msg = f"{cpu_temp:3.1f} C"
+            msg_size = image.string_size(msg, scale=self.font_scale)
+            img.draw_string(last[0] - msg_size.width() - 2, last[1] - msg_size.height() - 2, msg, image.COLOR_RED, scale=self.font_scale, thickness=2)
 
             img.draw_image(0, 0, self.img_back)
             img.draw_rect(self.string_cam_res_rect[0], self.string_cam_res_rect[1], self.string_cam_res_rect[2], self.string_cam_res_rect[3], image.COLOR_WHITE)

projects/app_benchmark/main.py

@@ -104,23 +104,34 @@ def run(self):
                     print("run item", self.runing_case_obj.name)
                     show, img = self.runing_case_obj.run()
                     print(f"run item {self.runing_case_obj.name} end")
-                    del self.runing_case_obj
-                    self.runing_case_obj = None
-                    gc.collect()
                     # show result
                     if show:
                         print("show result, press any key to exit")
                         self.showing_result = True
-                        self.disp.show(img)
-                        while self.showing_result:
-                            x, y, preesed = ts.read()
-                            if preesed:
-                                touch_pressed = True
-                            elif touch_pressed:
-                                touch_pressed = False
-                                self.showing_result = False
-                                break
-                            time.sleep_ms(50)
+                        idx = -1
+                        if isinstance(img, (list, tuple)):
+                            idx = 0
+                        while self.showing_result and not app.need_exit():
+                            img0 = img[idx] if idx>=0 else img
+                            self.disp.show(img0)
+                            save_path = f"/root/benchmark_result_{self.runing_case_obj.name}_{max(0, idx)}.png"
+                            print(f"save to {save_path}")
+                            img0.save(save_path)
+                            while self.showing_result:
+                                x, y, preesed = ts.read()
+                                if preesed:
+                                    touch_pressed = True
+                                elif touch_pressed:
+                                    touch_pressed = False
+                                    if idx >= 0 and idx < len(img):
+                                        idx += 1
+                                        continue
+                                    self.showing_result = False
+                                    break
+                                time.sleep_ms(50)
+                    del self.runing_case_obj
+                    self.runing_case_obj = None
+                    gc.collect()
 
 program = Program(disp)
 try:

projects/app_face_emotion/main.py

@@ -11,6 +11,17 @@
 def is_in_button(x, y, btn_pos):
     return x > btn_pos[0] and x < btn_pos[0] + btn_pos[2] and y > btn_pos[1] and y < btn_pos[1] + btn_pos[3]
 
+def get_back_btn_img(width):
+    ret_width = int(width * 0.1)
+    img_back = image.load("/maixapp/share/icon/ret.png")
+    w, h = (ret_width, img_back.height() * ret_width // img_back.width())
+    if w % 2 != 0:
+        w += 1
+    if h % 2 != 0:
+        h += 1
+    img_back = img_back.resize(w, h)
+    return img_back
+
 def main(disp):
     global curr_model
 
@@ -28,19 +39,25 @@ def main(disp):
     classifier = nn.Classifier(model=models[models_keys[curr_model]], dual_buff=False)
     cam = camera.Camera(detector.input_width(), detector.input_height(), detector.input_format())
 
+    font_scale = 2 if cam.height() >= 480 else 1.2
+    font_thickness = 2 if cam.height() >= 480 else 1
+    str_size = image.string_size("A", scale=font_scale, thickness=font_thickness)
+    str_w = str_size.width()
+    str_h = str_size.height()
+
     mode_pressed = False
     ts = touchscreen.TouchScreen()
-    img_back = image.load("/maixapp/share/icon/ret.png")
-    back_rect = [0, 0, 32, 32]
-    mode_rect = [0, cam.height() - 26, image.string_size(models_keys[curr_model]).width() + 6, 30]
+    img_back = get_back_btn_img(cam.width())
+    back_rect = [0, 0, img_back.width(), img_back.height()]
+    mode_rect = [0, cam.height() - int(str_h * 2), image.string_size(models_keys[curr_model], scale=font_scale, thickness=font_thickness).width() + 6, int(str_h * 2)]
     back_rect_disp = image.resize_map_pos(cam.width(), cam.height(), disp.width(), disp.height(), image.Fit.FIT_CONTAIN, back_rect[0], back_rect[1], back_rect[2], back_rect[3])
     mode_rect_disp = image.resize_map_pos(cam.width(), cam.height(), disp.width(), disp.height(), image.Fit.FIT_CONTAIN, mode_rect[0], mode_rect[1], mode_rect[2], mode_rect[3])
 
 
     # for draw result info
     max_labels_length = 0
     for label in classifier.labels:
-        size = image.string_size(label)
+        size = image.string_size(label, scale=font_scale, thickness=font_thickness)
         if size.width() > max_labels_length:
             max_labels_length = size.width()
 
@@ -73,18 +90,18 @@ def main(disp):
                 for j in range(len(classifier.labels)):
                     idx = res[j][0]
                     score = res[j][1]
-                    img.draw_string(0, img_std_first.height() + idx * 16, classifier.labels[idx], image.COLOR_WHITE)
-                    img.draw_rect(max_labels_length, int(img_std_first.height() + idx * 16), int(score * max_score_length), 8, image.COLOR_GREEN if score >= emotion_conf_th else image.COLOR_RED, -1)
-                    img.draw_string(int(max_labels_length + score * max_score_length + 2), int(img_std_first.height() + idx * 16), f"{score:.1f}", image.COLOR_RED)
+                    img.draw_string(0, img_std_first.height() + idx * str_h, classifier.labels[idx], image.COLOR_WHITE, font_scale, font_thickness)
+                    img.draw_rect(max_labels_length, int(img_std_first.height() + idx * str_h), int(score * max_score_length), 8, image.COLOR_GREEN if score >= emotion_conf_th else image.COLOR_RED, -font_thickness)
+                    img.draw_string(int(max_labels_length + score * max_score_length + 2), int(img_std_first.height() + idx * str_h), f"{score:.1f}", image.COLOR_RED, font_scale, font_thickness)
             # draw on all face
             color = image.COLOR_GREEN if res[0][1] >= emotion_conf_th else image.COLOR_RED
             obj = objs[idxes[i]]
             img.draw_rect(obj.x, obj.y, obj.w, obj.h, color, 1)
-            img.draw_string(obj.x, obj.y, f"{classifier.labels[res[0][0]]}: {res[0][1]:.1f}", color)
+            img.draw_string(obj.x, obj.y, f"{classifier.labels[res[0][0]]}: {res[0][1]:.1f}", color, font_scale, font_thickness)
 
         img.draw_image(0, 0, img_back)
         img.draw_rect(mode_rect[0], mode_rect[1], mode_rect[2], mode_rect[3], image.COLOR_WHITE)
-        img.draw_string(4, img.height() - 20, f"{models_keys[curr_model]}")
+        img.draw_string(4, img.height() - str_h  - str_h // 2, f"{models_keys[curr_model]}", scale=font_scale, thickness=font_thickness)
         disp.show(img)
         x, y, preesed = ts.read()
         if preesed:
@@ -97,8 +114,8 @@ def main(disp):
                 curr_model = (curr_model + 1) % len(models_keys)
                 msg = "switching model ..."
                 size = image.string_size(msg, scale=1.3)
-                img.draw_string((img.width() - size.width()) // 2, (img.height() - size.height())//2, msg, image.COLOR_RED, scale=1.3, thickness=-3)
-                img.draw_string((img.width() - size.width()) // 2, (img.height() - size.height())//2, msg, image.COLOR_WHITE, scale=1.3)
+                img.draw_string((img.width() - size.width()) // 2, (img.height() - size.height())//2, msg, image.COLOR_RED, font_scale, int(font_thickness*2))
+                img.draw_string((img.width() - size.width()) // 2, (img.height() - size.height())//2, msg, image.COLOR_WHITE, font_scale, font_thickness)
                 disp.show(img)
                 del detector
                 del landmarks_detector
@@ -111,6 +128,7 @@ def main(disp):
 except Exception:
     import traceback
     msg = traceback.format_exc()
+    print(msg)
     img = image.Image(disp.width(), disp.height())
     img.draw_string(0, 0, msg, image.COLOR_WHITE)
     disp.show(img)

projects/app_face_landmarks/main.py

@@ -37,6 +37,17 @@ def get_sub_landmarks(points, points_z, idxes):
 def is_in_button(x, y, btn_pos):
     return x > btn_pos[0] and x < btn_pos[0] + btn_pos[2] and y > btn_pos[1] and y < btn_pos[1] + btn_pos[3]
 
+def get_back_btn_img(width):
+    ret_width = int(width * 0.1)
+    img_back = image.load("/maixapp/share/icon/ret.png")
+    w, h = (ret_width, img_back.height() * ret_width // img_back.width())
+    if w % 2 != 0:
+        w += 1
+    if h % 2 != 0:
+        h += 1
+    img_back = img_back.resize(w, h)
+    return img_back
+
 def main(disp):
     global curr_sub
 
@@ -57,8 +68,8 @@ def main(disp):
     cam = camera.Camera(detector.input_width(), detector.input_height(), detector.input_format())
 
     ts = touchscreen.TouchScreen()
-    img_back = image.load("/maixapp/share/icon/ret.png")
-    back_rect = [0, 0, 32, 32]
+    img_back = get_back_btn_img(cam.width())
+    back_rect = [0, 0, img_back.width(), img_back.height()]
     mode_rect = [0, cam.height() - 26, 100, 30]
     back_rect_disp = image.resize_map_pos(cam.width(), cam.height(), disp.width(), disp.height(), image.Fit.FIT_CONTAIN, back_rect[0], back_rect[1], back_rect[2], back_rect[3])
     mode_rect_disp = image.resize_map_pos(cam.width(), cam.height(), disp.width(), disp.height(), image.Fit.FIT_CONTAIN, mode_rect[0], mode_rect[1], mode_rect[2], mode_rect[3])
@@ -102,6 +113,7 @@ def main(disp):
 except Exception:
     import traceback
     msg = traceback.format_exc()
+    print(msg)
     img = image.Image(disp.width(), disp.height())
     img.draw_string(0, 0, msg, image.COLOR_WHITE)
     disp.show(img)

projects/app_face_recognizer/main.py

@@ -4,6 +4,17 @@
 pressed_flag = [False, False, False]
 learn_id = 0
 
+def get_back_btn_img(width):
+    ret_width = int(width * 0.1)
+    img_back = image.load("/maixapp/share/icon/ret.png")
+    w, h = (ret_width, img_back.height() * ret_width // img_back.width())
+    if w % 2 != 0:
+        w += 1
+    if h % 2 != 0:
+        h += 1
+    img_back = img_back.resize(w, h)
+    return img_back
+
 def main(disp):
     global pressed_flag, learn_id
     img = image.Image(disp.width(), disp.height())
@@ -115,6 +126,7 @@ def on_touch(x, y, pressed):
 except Exception:
     import traceback
     msg = traceback.format_exc()
+    print(msg)
     img = image.Image(disp.width(), disp.height())
     img.draw_string(0, 0, msg, image.COLOR_WHITE)
     disp.show(img)

projects/app_face_tracking/main.py

@@ -22,21 +22,21 @@ def __init__(self, out_range:float, ignore_limit:float, path:str):
         self.roll = 0
         self.out_range = out_range
         self.ignore = ignore_limit
-        
+
         ### Self.w and self.h must be initialized.
         self.detector = nn.Retinaface(model=path)
         self.w = self.detector.input_width()
         self.h = self.detector.input_height()
         self.cam = camera.Camera(self.w, self.h)
         self.disp = display.Display()
-        
+
         ### The following section is used as an opt-out and normally you do not need to modify it.
         self.ts = touchscreen.TouchScreen()
         self.img_exit = image.load("./assets/exit.jpg").resize(40, 40)
         self.img_exit_touch = image.load("./assets/exit_touch.jpg").resize(40, 40)
         self.box = [0, 0, self.img_exit.width(), self.img_exit.height()]
         self.need_exit = False
-        
+
     def __check_touch_box(self, t, box, oft = 0):
         """This method is used for exiting and you normally do not need to modify or call it.
             You usually don't need to modify it.
@@ -45,7 +45,7 @@ def __check_touch_box(self, t, box, oft = 0):
             return True
         else:
             return False
-    
+
     def __exit_listener(self, img):
         """Exit case detection methods.
             It also draws the Exit button in the upper left corner.
@@ -60,7 +60,7 @@ def __exit_listener(self, img):
             self.need_exit = True
         else:
             img.draw_image(self.box[0], self.box[1], self.img_exit)
-            
+
     def is_need_exit(self):
         """Queries whether the exit button has been pressed.
             You usually don't need to modify it.
@@ -69,14 +69,14 @@ def is_need_exit(self):
             bool: Returns true if the exit button has been pressed, false otherwise.
         """
         return self.need_exit
-        
+
     def __get_target(self):
         """Get the coordinate value of the target.
             The behavior of this function needs to be customized.
         Returns:
             int, int: If no target is found, return -1,-1.
                       If the target is found, return the coordinate values x,y of the center point of the target.
-        """    
+        """
         ltime = time.ticks_ms()
         img = self.cam.read()               # Reads an image frame.
         objs = self.detector.detect(img, conf_th = 0.4, iou_th = 0.45)  # Recognition.
@@ -111,25 +111,25 @@ def get_target_err(self):
         if abs(self.roll) < self.out_range*self.ignore:
             self.roll = 0
         return self.pitch, self.roll
-          
+
 
 if __name__ == '__main__':
-    ROLL_PWM_PIN_NAME = "A17"   
+    ROLL_PWM_PIN_NAME = "A17"
     PITCH_PWM_PIN_NAME = "A16"
     init_pitch = 80             # init position, value: [0, 100], means minimum angle to maxmum angle of servo
     init_roll = 50              # 50 means middle
     PITCH_DUTY_MIN  = 3.5       # The minimum duty cycle corresponding to the range of motion of the y-axis servo.
     PITCH_DUTY_MAX  = 9.5       # Maximum duty cycle corresponding to the y-axis servo motion range.
     ROLL_DUTY_MIN   = 2.5       # Minimum duty cycle for x-axis servos.
     ROLL_DUTY_MAX   = 12.5      # Maxmum duty cycle for x-axis servos.
-    
+
     pitch_pid = [0.3, 0.0001, 0.0018, 0]    # [P I D I_max]
     roll_pid  = [0.3, 0.0001, 0.0018, 0]    # [P I D I_max]
     target_err_range = 10                   # target error output range, default [0, 10]
     target_ignore_limit = 0.08              # when target error < target_err_range*target_ignore_limit , set target error to 0
     pitch_reverse = False                   # reverse out value direction
     roll_reverse = True                     # reverse out value direction
-    
+
     target = Target(target_err_range, target_ignore_limit, MODEL)
     try:
         roll = servos.Servos(ROLL_PWM_PIN_NAME, init_roll, ROLL_DUTY_MIN, ROLL_DUTY_MAX)
@@ -145,17 +145,17 @@ def get_target_err(self):
             time.sleep(1)
             wait_time_s -= 1
         exit(-1)
-    
+
     pid_pitch = servos.PID(p=pitch_pid[0], i=pitch_pid[1], d=pitch_pid[2], imax=pitch_pid[3])
     pid_roll = servos.PID(p=roll_pid[0], i=roll_pid[1], d=roll_pid[2], imax=roll_pid[3])
     gimbal = servos.Gimbal(pitch, pid_pitch, roll, pid_roll)
-    
+
     total_uesd_time = 0
     total_fps = 0
     t0 = time.ticks_ms()
     while not target.is_need_exit():
         ltime = time.ticks_ms()
-        
+
         # get target error
         err_pitch, err_roll = target.get_target_err()
         # interval limit to >= 10ms

projects/app_hand_gesture_classifier/icon.png

@@ -1,7 +1,7 @@
 id: hand_landmarks
 name: Hand Landmarks
 name[zh]: 手关键点
-version: 1.0.1
+version: 1.0.2
 author: Neucrack@Sipeed
 icon: icon.png
 desc: Hand Landmarks detect hand keypoints