yolov4-tiny-3l issue resolved

hunglc007 · Mar 26, 2021 · 13e6adc · 13e6adc
1 parent 9f16748
commit 13e6adc
Show file tree

Hide file tree

Showing 7 changed files with 118 additions and 16 deletions.
diff --git a/convert_tflite.py b/convert_tflite.py
@@ -36,6 +36,7 @@ def save_tflite():
     converter.target_spec.supported_ops = [tf.lite.OpsSet.TFLITE_BUILTINS, tf.lite.OpsSet.SELECT_TF_OPS]
     converter.allow_custom_ops = True
   elif FLAGS.quantize_mode == 'int8':
+    converter.experimental_new_converter = False
     converter.target_spec.supported_ops = [tf.lite.OpsSet.TFLITE_BUILTINS_INT8]
     converter.optimizations = [tf.lite.Optimize.DEFAULT]
     converter.target_spec.supported_ops = [tf.lite.OpsSet.TFLITE_BUILTINS, tf.lite.OpsSet.SELECT_TF_OPS]

diff --git a/core/backbone.py b/core/backbone.py
@@ -146,6 +146,49 @@ def cspdarknet53_tiny(input_data):
 
     return route_1, input_data
 
+def cspdarknet53_tiny_3l(input_data):
+    input_data = common.convolutional(input_data, (3, 3, 3, 32), downsample=True)
+    input_data = common.convolutional(input_data, (3, 3, 32, 64), downsample=True)
+    input_data = common.convolutional(input_data, (3, 3, 64, 64))
+
+    route = input_data
+    input_data = common.route_group(input_data, 2, 1)
+    input_data = common.convolutional(input_data, (3, 3, 32, 32))
+    route_1 = input_data
+    input_data = common.convolutional(input_data, (3, 3, 32, 32))
+    input_data = tf.concat([input_data, route_1], axis=-1)
+    input_data = common.convolutional(input_data, (1, 1, 64, 64))
+    input_data = tf.concat([route, input_data], axis=-1)
+    input_data = tf.keras.layers.MaxPool2D(2, 2, 'same')(input_data)
+
+    input_data = common.convolutional(input_data, (3, 3, 128, 128))
+    route = input_data
+    input_data = common.route_group(input_data, 2, 1)
+    input_data = common.convolutional(input_data, (3, 3, 64, 64))
+    route_1 = input_data
+    input_data = common.convolutional(input_data, (3, 3, 64, 64))
+    input_data = tf.concat([input_data, route_1], axis=-1)
+    input_data = common.convolutional(input_data, (1, 1, 128, 128))
+    route_2 = input_data
+    input_data = tf.concat([route, input_data], axis=-1)
+    input_data = tf.keras.layers.MaxPool2D(2, 2, 'same')(input_data)
+
+    input_data = common.convolutional(input_data, (3, 3, 256, 256))
+    route = input_data
+    input_data = common.route_group(input_data, 2, 1)
+    input_data = common.convolutional(input_data, (3, 3, 128, 128))
+    route_1 = input_data
+    input_data = common.convolutional(input_data, (3, 3, 128, 128))
+    input_data = tf.concat([input_data, route_1], axis=-1)
+    input_data = common.convolutional(input_data, (1, 1, 256, 256))
+    route_1 = input_data
+    input_data = tf.concat([route, input_data], axis=-1)
+    input_data = tf.keras.layers.MaxPool2D(2, 2, 'same')(input_data)
+
+    input_data = common.convolutional(input_data, (3, 3, 512, 512))
+
+    return route_1, route_2, input_data
+
 def darknet53_tiny(input_data):
     input_data = common.convolutional(input_data, (3, 3, 3, 16))
     input_data = tf.keras.layers.MaxPool2D(2, 2, 'same')(input_data)

diff --git a/core/config.py b/core/config.py
@@ -11,14 +11,18 @@
 # YOLO options
 __C.YOLO                      = edict()
 
-__C.YOLO.CLASSES              = "./data/classes/coco.names"
+__C.YOLO.CLASSES              = "./data/classes/obj.names"
 __C.YOLO.ANCHORS              = [12,16, 19,36, 40,28, 36,75, 76,55, 72,146, 142,110, 192,243, 459,401]
 __C.YOLO.ANCHORS_V3           = [10,13, 16,30, 33,23, 30,61, 62,45, 59,119, 116,90, 156,198, 373,326]
-__C.YOLO.ANCHORS_TINY         = [23,27, 37,58, 81,82, 81,82, 135,169, 344,319]
+# __C.YOLO.ANCHORS_TINY         = [23,27, 37,58, 81,82, 81,82, 135,169, 344,319]
+__C.YOLO.ANCHORS_TINY         = [10,14,  23,27,  37,58,  81,82,  135,169,  344,319]
+__C.YOLO.ANCHORS_TINY_3l      = [12,16, 19,36, 40,28, 36,75, 76,55, 72,146, 142,110, 192,243, 459,401]
 __C.YOLO.STRIDES              = [8, 16, 32]
 __C.YOLO.STRIDES_TINY         = [16, 32]
+__C.YOLO.STRIDES_TINY_3l      = [8, 16, 32]
 __C.YOLO.XYSCALE              = [1.2, 1.1, 1.05]
 __C.YOLO.XYSCALE_TINY         = [1.05, 1.05]
+__C.YOLO.XYSCALE_TINY_3l      = [1.05, 1.05, 1.05]
 __C.YOLO.ANCHOR_PER_SCALE     = 3
 __C.YOLO.IOU_LOSS_THRESH      = 0.5
 

diff --git a/core/utils.py b/core/utils.py
@@ -23,6 +23,9 @@ def load_weights(model, weights_file, model_name='yolov4', is_tiny=False):
         if model_name == 'yolov3':
             layer_size = 13
             output_pos = [9, 12]
+        elif model_name == 'yolov4-tiny-3l':
+            layer_size = 24
+            output_pos = [17, 20, 23]
         else:
             layer_size = 21
             output_pos = [17, 20]
@@ -81,24 +84,32 @@ def read_class_names(class_file_name):
 
 def load_config(FLAGS):
     if FLAGS.tiny:
-        STRIDES = np.array(cfg.YOLO.STRIDES_TINY)
-        ANCHORS = get_anchors(cfg.YOLO.ANCHORS_TINY, FLAGS.tiny)
-        XYSCALE = cfg.YOLO.XYSCALE_TINY if FLAGS.model == 'yolov4' else [1, 1]
+        if FLAGS.model == 'yolov4-tiny-3l':
+            STRIDES = np.array(cfg.YOLO.STRIDES_TINY_3l)
+            ANCHORS = get_anchors(cfg.YOLO.ANCHORS_TINY_3l, FLAGS.tiny, FLAGS.model)
+            XYSCALE = cfg.YOLO.XYSCALE_TINY_3l
+        else:
+            STRIDES = np.array(cfg.YOLO.STRIDES_TINY)
+            ANCHORS = get_anchors(cfg.YOLO.ANCHORS_TINY, FLAGS.tiny, FLAGS.model)
+            XYSCALE = cfg.YOLO.XYSCALE_TINY if FLAGS.model == 'yolov4' else [1, 1]
     else:
         STRIDES = np.array(cfg.YOLO.STRIDES)
         if FLAGS.model == 'yolov4':
-            ANCHORS = get_anchors(cfg.YOLO.ANCHORS, FLAGS.tiny)
+            ANCHORS = get_anchors(cfg.YOLO.ANCHORS, FLAGS.tiny, FLAGS.model)
         elif FLAGS.model == 'yolov3':
-            ANCHORS = get_anchors(cfg.YOLO.ANCHORS_V3, FLAGS.tiny)
+            ANCHORS = get_anchors(cfg.YOLO.ANCHORS_V3, FLAGS.tiny, FLAGS.model)
         XYSCALE = cfg.YOLO.XYSCALE if FLAGS.model == 'yolov4' else [1, 1, 1]
     NUM_CLASS = len(read_class_names(cfg.YOLO.CLASSES))
 
     return STRIDES, ANCHORS, NUM_CLASS, XYSCALE
 
-def get_anchors(anchors_path, tiny=False):
+def get_anchors(anchors_path, tiny=False, model='yolov4'):
     anchors = np.array(anchors_path)
     if tiny:
-        return anchors.reshape(2, 3, 2)
+        if model == 'yolov4-tiny-3l':
+            return anchors.reshape(3, 3, 2)
+        else:
+            return anchors.reshape(2, 3, 2)
     else:
         return anchors.reshape(3, 3, 2)
 

diff --git a/core/yolov4.py b/core/yolov4.py
@@ -18,6 +18,13 @@ def YOLO(input_layer, NUM_CLASS, model='yolov4', is_tiny=False):
     if is_tiny:
         if model == 'yolov4':
             return YOLOv4_tiny(input_layer, NUM_CLASS)
+        elif model == 'yolov4-tiny-3l':
+            print("#"*100)
+            print("\n"*5)
+            print("loading tiny-3l")
+            print("\n"*5)
+            print("#"*100)
+            return YOLOv4_tiny_3l(input_layer, NUM_CLASS)
         elif model == 'yolov3':
             return YOLOv3_tiny(input_layer, NUM_CLASS)
     else:
@@ -143,6 +150,29 @@ def YOLOv4_tiny(input_layer, NUM_CLASS):
 
     return [conv_mbbox, conv_lbbox]
 
+def YOLOv4_tiny_3l(input_layer, NUM_CLASS):
+    route_1, route_2, conv = backbone.cspdarknet53_tiny_3l(input_layer)
+
+    conv = common.convolutional(conv, (1, 1, 512, 256))
+
+    conv_lobj_branch = common.convolutional(conv, (3, 3, 256, 512))
+    conv_lbbox = common.convolutional(conv_lobj_branch, (1, 1, 512, 3 * (NUM_CLASS + 5)), activate=False, bn=False)
+
+    conv = common.convolutional(conv, (1, 1, 256, 128))
+    conv = common.upsample(conv)
+    conv = tf.concat([conv, route_1], axis=-1)
+
+    conv = common.convolutional(conv, (3, 3, 384, 256))
+    conv_mbbox = common.convolutional(conv, (1, 1, 256, 3 * (NUM_CLASS + 5)), activate=False, bn=False)
+
+    conv = common.convolutional(conv, (1, 1, 256, 64))
+    conv = common.upsample(conv)
+    conv = tf.concat([conv, route_2], axis=-1)
+
+    conv_sobj_branch = common.convolutional(conv, (3, 3, 192, 128))
+    conv_sbbox = common.convolutional(conv_sobj_branch, (1, 1, 128, 3 * (NUM_CLASS + 5)), activate=False, bn=False)
+    return [conv_sbbox, conv_mbbox, conv_lbbox]
+
 def YOLOv3_tiny(input_layer, NUM_CLASS):
     route_1, conv = backbone.darknet53_tiny(input_layer)
 

diff --git a/data/classes/obj.names b/data/classes/obj.names
@@ -0,0 +1 @@
+license_plate
diff --git a/save_model.py b/save_model.py
@@ -21,13 +21,25 @@ def save_tf():
   bbox_tensors = []
   prob_tensors = []
   if FLAGS.tiny:
-    for i, fm in enumerate(feature_maps):
-      if i == 0:
-        output_tensors = decode(fm, FLAGS.input_size // 16, NUM_CLASS, STRIDES, ANCHORS, i, XYSCALE, FLAGS.framework)
-      else:
-        output_tensors = decode(fm, FLAGS.input_size // 32, NUM_CLASS, STRIDES, ANCHORS, i, XYSCALE, FLAGS.framework)
-      bbox_tensors.append(output_tensors[0])
-      prob_tensors.append(output_tensors[1])
+    if FLAGS.model ==  'yolov4-tiny-3l':
+      for i, fm in enumerate(feature_maps):
+        # import pdb; pdb.set_trace()
+        if i == 0:
+          output_tensors = decode(fm, FLAGS.input_size // 8, NUM_CLASS, STRIDES, ANCHORS, i, XYSCALE, FLAGS.framework)
+        elif i == 1:
+          output_tensors = decode(fm, FLAGS.input_size // 16, NUM_CLASS, STRIDES, ANCHORS, i, XYSCALE, FLAGS.framework)
+        else:
+          output_tensors = decode(fm, FLAGS.input_size // 32, NUM_CLASS, STRIDES, ANCHORS, i, XYSCALE, FLAGS.framework)
+        bbox_tensors.append(output_tensors[0])
+        prob_tensors.append(output_tensors[1])    
+    else:
+      for i, fm in enumerate(feature_maps):
+        if i == 0:
+          output_tensors = decode(fm, FLAGS.input_size // 16, NUM_CLASS, STRIDES, ANCHORS, i, XYSCALE, FLAGS.framework)
+        else:
+          output_tensors = decode(fm, FLAGS.input_size // 32, NUM_CLASS, STRIDES, ANCHORS, i, XYSCALE, FLAGS.framework)
+        bbox_tensors.append(output_tensors[0])
+        prob_tensors.append(output_tensors[1])  
   else:
     for i, fm in enumerate(feature_maps):
       if i == 0: