code cleanup

README.md

@@ -142,6 +142,7 @@ Examples of augmentation are given in [https://github.com/matterport/Mask_RCNN](
 
 #### Training Loss
 
+The model was trained using pretrained CoCo weights.
 Instead of a single training loop, it was trained multiple times in smaller epochs to observe change in the loss with changes in parameters and to avoid overfitting. As the data was less, the network used to saturate quickly and required more augmentations to proceed. Also i did not wanted to go overboard on the augmentation so was observing which one works best. Below are the logs of final training setting with the above given augmentation.   
 
 | heads Epoch | all Epoch | loss 							| val_loss 								|

inference.py

@@ -12,9 +12,8 @@
 import matplotlib.pyplot as plt
 import skimage.io
 import time
-# from imgaug import augmenters as iaa
 
-# os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
+os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
 
 ROOT_DIR = os.path.abspath("./")
 MODEL_DIR = os.path.join('./', "logs")
@@ -31,7 +30,7 @@
 
 
 
-class ShapesConfig(Config):
+class LyftChallengeConfig(Config):
     """Configuration for training on the toy shapes dataset.
     Derives from the base Config class and overrides values specific
     to the toy shapes dataset.
@@ -68,10 +67,10 @@ class ShapesConfig(Config):
 
 
 
-config = ShapesConfig()
+config = LyftChallengeConfig()
 config.display()
 
-class InferenceConfig(ShapesConfig):
+class InferenceConfig(LyftChallengeConfig):
     GPU_COUNT = 1
     IMAGES_PER_GPU = 1
 
@@ -99,9 +98,6 @@ class InferenceConfig(ShapesConfig):
 
 def segment_images(original_image):
     results = model.detect([original_image], verbose=0)
-    # print ("-"*80)
-    # print(len(results))
-    # print ("-"*80)
     r = results[0]
     f_mask = r['masks']
     f_class = r["class_ids"]
@@ -123,34 +119,6 @@ def segment_images(original_image):
         final_img = cv2.addWeighted(final_img, 1, mask1.astype(np.uint8), 1, 0)
     return final_img
 
-def segment_images_batch(original_images):
-    results = model.detect(original_images, verbose=0)
-    print ("-"*80)
-    print(len(results))
-    print ("-"*80)
-    images = []
-    for idx,res in enumerate(results):
-        r = res
-        f_mask = r['masks']
-        f_class = r["class_ids"]
-
-
-        no_ch = f_mask.shape[2]
-        final_img = np.copy(original_images[idx,:,:,:])
-        for ch in range(no_ch):
-
-            _id = f_class[ch]
-            if _id==1: 
-                color_id=0
-            else:
-                color_id=1
-            mask_1 = f_mask[:,:,ch]
-            mask1 = np.dstack([mask_1*colors[color_id][0],
-                                mask_1*colors[color_id][1],
-                                mask_1*colors[color_id][2]])
-            final_img = cv2.addWeighted(final_img, 1, mask1.astype(np.uint8), 1, 0)
-            images.append(final_img)
-    return np.dstack(images)
 
 import sys, skvideo.io, json, base64
 

test_inference.py

@@ -8,11 +8,11 @@
 
 os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
 
-ROOT_DIR = os.path.abspath("./Lyft_challenge")
-MODEL_DIR = os.path.join('./Lyft_challenge', "logs")
+ROOT_DIR = os.path.abspath("./")
+MODEL_DIR = os.path.join('./', "logs")
 
 sys.path.append(ROOT_DIR)  # To find local version of the library
-sys.path.append(os.path.join(os.getcwd(),"./Lyft_challenge/Mask_RCNN/"))
+sys.path.append(os.path.join(os.getcwd(),"./Mask_RCNN/"))
 
 from mrcnn.config import Config
 from mrcnn import utils
@@ -21,13 +21,13 @@
 from mrcnn.model import log
 
 
-class ShapesConfig(Config):
+class LyftChallengeConfig(Config):
     """Configuration for training on the toy shapes dataset.
     Derives from the base Config class and overrides values specific
     to the toy shapes dataset.
     """
     # Give the configuration a recognizable name
-    NAME = "shapes"
+    NAME = "lyft_perception_challenge"
 
     # Train on 1 GPU and 8 images per GPU. We can put multiple images on each
     # GPU because the images are small. Batch size is 8 (GPUs * images/GPU).
@@ -57,11 +57,11 @@ class ShapesConfig(Config):
     VALIDATION_STEPS = 5
 
 
-config = ShapesConfig()
+config = LyftChallengeConfig()
 # config.display()
 
 
-class InferenceConfig(ShapesConfig):
+class InferenceConfig(LyftChallengeConfig):
     GPU_COUNT = 1
     IMAGES_PER_GPU = 1
 
@@ -77,7 +77,7 @@ class InferenceConfig(ShapesConfig):
                           config=inference_config,
                           model_dir=MODEL_DIR)
 
-model_path = os.path.join('./Lyft_challenge', "mask_rcnn_lyft.h5")
+model_path = os.path.join('./', "mask_rcnn_lyft.h5")
 assert model_path != "", "Provide path to trained weights"
 # print("Loading weights from ", model_path)
 model.load_weights(model_path, by_name=True)
@@ -101,12 +101,6 @@ def segment_image(image_frame):
 
     return car_mask,road_mask
 
-# Define encoder function
-# def encode(array):
-#     pil_img = Image.fromarray(array)
-#     buff = BytesIO()
-#     pil_img.save(buff, format="PNG")
-#     return base64.b64encode(buff.getvalue()).decode("utf-8")
 
 def encode(array):
     retval, buffer = cv2.imencode('.png', array)
@@ -121,13 +115,7 @@ def encode(array):
 
 for rgb_frame in video:
 
-    # Grab red channel  
-    # red = rgb_frame[:,:,0]    
-    # Look for red cars :)
-    # 
 
-    # Look for road :)
-    # 
     car_mask,road_mask = segment_image(rgb_frame)
     binary_car_result = car_mask*1
     binary_road_result = road_mask*1