computer_vision_utils

Everything that I code more than twice during my PhD will end up here.

bounding boxes

• bbox_helper.py

  Defined class Rectangle which should help in all situations that involve handling of bounding boxes.

  class Rectangle(x_min, y_min, x_max, y_max)
      Methods:
          intersect_with(self, rect)
          resize_sides(self, ratio, bounds=None)
          draw(self, frame, color=255, thickness=1)
          get_binary_mask(self, mask_shape)
      Properties:
          tl_corner(self)
          br_corner(self)
          coords(self)
          area(self)
  

stitching

• stitching.py

  stitch_together(input_images, layout, resize_dim=None, off_x=None, off_y=None, bg_color=(0, 0, 0)):

  Stitch together N input images into a bigger frame, using a grid layout. Input images can be either color or grayscale, but must all have the same size. Background color is black by default, but it can be changed with bg_color parameter.

containers

• list_helper.py

  dump_list(input_list, file_path)

  Dump list to file, either in "txt" or binary ("pickle") mode. Dump mode is chosen accordingly to "file_path" extension.

  load_list(file_path)

  Load list from file, either in "txt" or binary ("pickle") mode. Load mode is chosen accordingly to "file_path" extension.

  split_into_chunks(list_in, max_elements, shuffle=False)

  Split a list a variable number of chunks of at most "max_elements" each.

filesystem

• filesystem.py

  get_file_list_recursively(top_directory, allowed_extension=[])

  Get list of full paths of all files found under root directory "top_directory". If a list of allowed file extensions is provided, files are filtered according to this list.

io

• io_helper.py

  read_image(path, channels_first, color=True, dtype=np.float32, resize_dim=None)

  Reads an image from "path" and returns respecting the self explanatory parameters

  write_image(img_path, img, channels_first=False, color_mode='RGB', resize_dim=None)

  Writes an image into "img_path" file. If color, you must specify whether the color dimension is the first one or the last one with "channels_first", and the "color_mode" as well. Optionally one can resize the image.

  normalize(img)

  Normalizes an image between 0 and 255 and returns it as uint8.

tensor_manipulation

• tensor_manipulation.py

  resize_tensor(tensor, new_size)

  Resizes a numeri tensor having shape (channels, h, w) into the new size (channels, new_h, new_w). Each channel is resized indipendently (good for feature maps).

  crop_tensor(tensor, indexes)

  Crops a numeric tensor having shape (channels, h, w) according to indexes in the form (h1,h2,w1,w2).