PEP8 standard

README.md

@@ -1,5 +1,18 @@
 # PY-GOTURN
 
+### News
+I'm back on working on this codebase again, can expect the following
+changes on the development branch (goturn-dev).
+
+- [] Code clean up
+
+- [] Try to add much better interface for easy debugging, to understand different parts of
+the code
+
+- [] Documentation or blog on the code.
+
+**NOTE:** If you wish to contribute, please ping me
+
 This is the **python** implementation of **GOTURN: Generic Object Tracking Using Regression Networks.**
 
 **[Learning to Track at 100 FPS with Deep Regression Networks](http://davheld.github.io/GOTURN/GOTURN.html)**,

goturn/helper/BoundingBox.py

@@ -5,6 +5,7 @@
 
 from ..helper.helper import sample_exp_two_sides, sample_rand_uniform
 
+
 class BoundingBox:
     """Docstring for BoundingBox. """
 
@@ -24,14 +25,14 @@ def get_center_x(self):
         :returns: TODO
 
         """
-        return (self.x1 + self.x2)/2.
+        return (self.x1 + self.x2) / 2.
 
     def get_center_y(self):
         """TODO: Docstring for get_center_y.
         :returns: TODO
 
         """
-        return (self.y1 + self.y2)/2.
+        return (self.y1 + self.y2) / 2.
 
     def compute_output_height(self):
         """TODO: Docstring for compute_output_height.
@@ -40,7 +41,7 @@ def compute_output_height(self):
         """
         bbox_height = self.y2 - self.y1
         output_height = self.kContextFactor * bbox_height
-        
+
         return max(1.0, output_height)
 
     def compute_output_width(self):
@@ -50,7 +51,7 @@ def compute_output_width(self):
         """
         bbox_width = self.x2 - self.x1
         output_width = self.kContextFactor * bbox_width
-        
+
         return max(1.0, output_width)
 
     def edge_spacing_x(self):
@@ -171,7 +172,6 @@ def shift(self, image, lambda_scale_frac, lambda_shift_frac, min_scale, max_scal
             new_width = max(1.0, min((image.shape[1] - 1), new_width))
             num_tries_width = num_tries_width + 1
 
-
         new_height = -1
         num_tries_height = 0
         while ((new_height < 0) or (new_height > image.shape[0] - 1)) and (num_tries_height < kMaxNumTries):
@@ -181,11 +181,10 @@ def shift(self, image, lambda_scale_frac, lambda_shift_frac, min_scale, max_scal
                 rand_num = sample_rand_uniform()
                 height_scale_factor = rand_num * (max_scale - min_scale) + min_scale
 
-            new_height = height * ( 1 + height_scale_factor )
+            new_height = height * (1 + height_scale_factor)
             new_height = max(1.0, min((image.shape[0] - 1), new_height))
             num_tries_height = num_tries_height + 1
 
-
         first_time_x = True
         new_center_x = -1
         num_tries_x = 0

goturn/helper/config.py

@@ -1,9 +1,11 @@
-# Date: Saturday 29 July 2017 
+# Date: Saturday 29 July 2017
 # Email: [email protected]
 # Name: Nrupatunga
 # Description: bounding box class
 
+import sys
 
 CAFFE_PATH = '/usr/local/caffe/python'
+sys.path.insert(0, CAFFE_PATH)
 DEBUG = False
-OPENCV_VERSION=2
+OPENCV_VERSION = 2

goturn/helper/helper.py

@@ -8,6 +8,7 @@
 
 RAND_MAX = 2147483647
 
+
 def sample_rand_uniform():
     """TODO: Docstring for sample_rand_uniform.
 
@@ -17,6 +18,7 @@ def sample_rand_uniform():
     """
     return (random.randint(0, RAND_MAX) + 1) * 1.0 / (RAND_MAX + 2)
 
+
 def sample_exp_two_sides(lambda_):
     """TODO: Docstring for sample_exp_two_sides.
     :returns: TODO

goturn/helper/image_proc.py

@@ -4,7 +4,6 @@
 # Description: Image processing functions
 
 import math
-import cv2
 import numpy as np
 from ..helper.BoundingBox import BoundingBox
 
@@ -20,7 +19,7 @@ def cropPadImage(bbox_tight, image):
     roi_width = min(image.shape[1], max(1.0, math.ceil(pad_image_location.x2 - pad_image_location.x1)))
     roi_height = min(image.shape[0], max(1.0, math.ceil(pad_image_location.y2 - pad_image_location.y1)))
 
-    err = 0.000000001 # To take care of floating point arithmetic errors
+    err = 0.000000001  # To take care of floating point arithmetic errors
     cropped_image = image[int(roi_bottom + err):int(roi_bottom + roi_height), int(roi_left + err):int(roi_left + roi_width)]
     output_width = max(math.ceil(bbox_tight.compute_output_width()), roi_width)
     output_height = max(math.ceil(bbox_tight.compute_output_height()), roi_height)
@@ -31,11 +30,11 @@ def cropPadImage(bbox_tight, image):
 
     edge_spacing_x = min(bbox_tight.edge_spacing_x(), (image.shape[1] - 1))
     edge_spacing_y = min(bbox_tight.edge_spacing_y(), (image.shape[0] - 1))
-    
+
     # if output_image[int(edge_spacing_y):int(edge_spacing_y) + cropped_image.shape[0], int(edge_spacing_x):int(edge_spacing_x) + cropped_image.shape[1]].shape != cropped_image.shape :
-        # import pdb
-        # pdb.set_trace()
-        # print('debug')
+    # import pdb
+    # pdb.set_trace()
+    # print('debug')
 
     # rounding should be done to match the width and height
     output_image[int(edge_spacing_y):int(edge_spacing_y) + cropped_image.shape[0], int(edge_spacing_x):int(edge_spacing_x) + cropped_image.shape[1]] = cropped_image
@@ -74,5 +73,5 @@ def computeCropPadImageLocation(bbox_tight, image):
 
     # Padded image location in the original image
     objPadImageLocation = BoundingBox(roi_left, roi_bottom, roi_left + roi_width, roi_bottom + roi_height)
-   
+
     return objPadImageLocation

goturn/loader/annotation.py

@@ -1,11 +1,11 @@
-# Date: Nrupatunga: Tuesday 04 July 2017 
+# Date: Nrupatunga: Tuesday 04 July 2017
 # Email: [email protected]
 # Name: Nrupatunga
 # Description: annotations
 
-import sys
 from ..helper.BoundingBox import BoundingBox
 
+
 class annotation:
 
     """Docstring for annotation. """

goturn/loader/loader_alov.py

@@ -3,15 +3,10 @@
 # Name: Nrupatunga
 # Description: loading Alov dataset
 
-import sys
-# sys.path.append('../logger/')
-# sys.path.append('../helper/')
 import os
 import glob
-from annotation import annotation
 from video import video
 from video import frame
-import xml.etree.ElementTree as ET
 from ..logger.logger import setup_logger
 from ..helper.BoundingBox import BoundingBox
 
@@ -37,14 +32,11 @@ def loaderAlov(self):
 
         """
         logger = self.logger
-        alov_folder = self.alov_folder
         alov_subdirs = sorted(self.find_subfolders(self.annotations_folder))
-        num_annotations = 0
-        dict_list_of_annotations = {}
 
         for i, alov_sub_folder in enumerate(alov_subdirs):
             annotations_files = sorted(glob.glob(os.path.join(self.annotations_folder, alov_sub_folder, '*.ann')))
-            logger.info('Loading {:>3} of {:>3} - annotation file from folder = {:>4}'.format(i+1, len(alov_subdirs), alov_sub_folder))
+            logger.info('Loading {:>3} of {:>3} - annotation file from folder = {:>4}'.format(i + 1, len(alov_subdirs), alov_sub_folder))
 
             for ann in annotations_files:
                 self.load_annotation_file(alov_sub_folder, ann)
@@ -68,7 +60,7 @@ def load_annotation_file(self, alov_sub_folder, annotation_file):
 
         with open(annotation_file, 'r') as f:
             data = f.read().rstrip().split('\n')
-            for bb in data: 
+            for bb in data:
                 frame_num, ax, ay, bx, by, cx, cy, dx, dy = bb.split()
                 frame_num, ax, ay, bx, by, cx, cy, dx, dy = int(frame_num), float(ax), float(ay), float(bx), float(by), float(cx), float(cy), float(dx), float(dy)
 
@@ -98,7 +90,6 @@ def get_videos(self, isTrain=True, val_ratio=0.2):
         num_categories = len(self.category)
         category = self.category
         keys = sorted(category.keys())
-        count = 0
         for i in range(num_categories):
             category_video = category[keys[i]]
             num_videos = len(category_video)
@@ -124,7 +115,7 @@ def get_videos(self, isTrain=True, val_ratio=0.2):
 
         return videos
 
-        
+
 if '__main__' == __name__:
     logger = setup_logger(logfile=None)
     objLoaderAlov = loader_alov('/media/nrupatunga/data/datasets/VOT-extract/images/', '/media/nrupatunga/data/datasets/VOT-extract/gt/', logger)

goturn/loader/loader_imagenet.py

@@ -4,7 +4,6 @@
 # Description: loading Imagenet dataset
 
 from __future__ import print_function
-import sys
 import os
 import cv2
 import glob
@@ -13,9 +12,9 @@
 from ..logger.logger import setup_logger
 from ..helper import config
 
-
 kMaxRatio = 0.66
 
+
 class loader_imagenet:
 
     """Docstring for loader_imagenetdet. """
@@ -35,14 +34,13 @@ def loaderImageNetDet(self):
 
         """
         logger = self.logger
-        imagenet_folder = self.imagenet_folder
         imagenet_subdirs = sorted(self.find_subfolders(self.annotations_folder))
         num_annotations = 0
         list_of_annotations_out = []
 
         for i, imgnet_sub_folder in enumerate(imagenet_subdirs):
             annotations_files = sorted(glob.glob(os.path.join(self.annotations_folder, imgnet_sub_folder, '*.xml')))
-            logger.info('Loading {}/{} - annotation file from folder = {}'.format(i+1, len(imagenet_subdirs), imgnet_sub_folder))
+            logger.info('Loading {}/{} - annotation file from folder = {}'.format(i + 1, len(imagenet_subdirs), imgnet_sub_folder))
             for ann in annotations_files:
                 list_of_annotations, num_ann_curr = self.load_annotation_file(ann)
                 num_annotations = num_annotations + num_ann_curr
@@ -57,7 +55,6 @@ def loaderImageNetDet(self):
         self.num_annotations = num_annotations
 
         return list_of_annotations_out
-
 
     def find_subfolders(self, imagenet_folder):
         """TODO: Docstring for find_subfolders.
@@ -118,7 +115,6 @@ def load_annotation(self, image_num, annotation_num):
         list_annotations = images[image_num]
         random_ann = list_annotations[annotation_num]
 
-
         img_path = os.path.join(self.imagenet_folder, random_ann.image_path + '.JPEG')
 
         if config.DEBUG:
@@ -130,7 +126,6 @@ def load_annotation(self, image_num, annotation_num):
             random_ann.disp_height = 375
             random_ann.disp_width = 500
 
-
         image = cv2.imread(img_path)
 
         img_height = image.shape[0]

goturn/loader/video.py

@@ -5,9 +5,11 @@
 
 import cv2
 
+
 class frame:
 
     """Docstring for frame. """
+
     def __init__(self, frame_num, bbox):
         """TODO: to be defined1. """
         self.frame_num = frame_num
@@ -33,12 +35,11 @@ def load_annotation(self, annotation_index):
         frame_num = ann_frame.frame_num
         bbox = ann_frame.bbox
 
-        video_path = self.video_path
-        image_files =  self.all_frames
+        image_files = self.all_frames
 
         assert(len(image_files) > 0)
         assert(frame_num < len(image_files))
-        
+
         image = cv2.imread(image_files[frame_num])
         return frame_num, image, bbox
 

goturn/network/regressor.py

@@ -170,4 +170,4 @@ def estimate(self, curr_search_region, target_region):
         net.forward()
         bbox_estimate = net.blobs['fc8'].data
 
-        return bbox_estimate
+        return bbox_estimate

goturn/network/regressor_train.py

@@ -9,8 +9,9 @@
 from ..helper import config
 sys.path.insert(0, config.CAFFE_PATH)
 import caffe
-from visdom import Visdom
-viz = Visdom()
+# __import__('pdb').set_trace()
+# from visdom import Visdom
+# viz = Visdom()
 
 class regressor_train:
 

goturn/test/show_tracker_vot.py

@@ -10,21 +10,20 @@
 from ..loader.loader_vot import loader_vot
 from ..tracker.tracker import tracker
 from ..tracker.tracker_manager import tracker_manager
-from ..loader.video import video
 
 setproctitle.setproctitle('SHOW_TRACKER_VOT')
 logger = setup_logger(logfile=None)
 
 ap = argparse.ArgumentParser()
-ap.add_argument("-p", "--prototxt", required = True, help = "Path to the prototxt")
-ap.add_argument("-m", "--model", required = True, help = "Path to the model")
-ap.add_argument("-v", "--input", required = True, help = "Path to the vot directory")
-ap.add_argument("-g", "--gpuID", required = True, help = "gpu to use")
+ap.add_argument("-p", "--prototxt", required=True, help="Path to the prototxt")
+ap.add_argument("-m", "--model", required=True, help="Path to the model")
+ap.add_argument("-v", "--input", required=True, help="Path to the vot directory")
+ap.add_argument("-g", "--gpuID", required=True, help="gpu to use")
 args = vars(ap.parse_args())
 
 do_train = False
 objRegressor = regressor(args['prototxt'], args['model'], args['gpuID'], 1, do_train, logger)
-objTracker = tracker(False, logger) # Currently no idea why this class is needed, eventually we shall figure it out
+objTracker = tracker(False, logger)  # Currently no idea why this class is needed, eventually we shall figure it out
 objLoaderVot = loader_vot(args['input'], logger)
 videos = objLoaderVot.get_videos()
 objTrackerVis = tracker_manager(videos, objRegressor, objTracker, logger)