Bounding Box for objects using Pytorch

This repository is built to understand object detection and bounding box from scratch. Here we implemet a simple geomentrical shape detection 🔹 🔺 and construct a bounding box ✏️ over the object. This is just a toy example.

Refrence:Pytorch
REF paper

Requirements

 1. Pytorch
 2. Pycairo

Usage

Prepare Training Dataset

To generate random Rectangles

bboxes = np.zeros((num_imgs, num_objects, 4))
imgs = np.zeros((num_imgs, img_size, img_size))  # set background to 0
for i_img in range(num_imgs):
    for i_object in range(num_objects):
        w, h = np.random.randint(min_object_size, max_object_size, size=2)
        x = np.random.randint(0, img_size - w)
        y = np.random.randint(0, img_size - h)
        imgs[i_img, x:x+w, y:y+h] = 1.  # set rectangle to 1
        bboxes[i_img, i_object] = [x, y, w, h]

Similarly, we generate images of multiple objects with their bounding boxes

Architecture

Single bounding box

LinearRegressionModel(
  (linear1): Linear(in_features=64, out_features=200, bias=True)
  (linear2): Linear(in_features=200, out_features=4, bias=True)
)

Multiple bounding box and object classification

----------------------------------------------------------------
        Layer (type)               Output Shape         Param #
================================================================
            Conv2d-1        [-1, 32L, 18L, 18L]             896
              ReLU-2        [-1, 32L, 18L, 18L]               0
         MaxPool2d-3          [-1, 32L, 9L, 9L]               0
            Conv2d-4        [-1, 64L, 11L, 11L]           18496
              ReLU-5        [-1, 64L, 11L, 11L]               0
         MaxPool2d-6          [-1, 64L, 5L, 5L]               0
            Conv2d-7         [-1, 128L, 7L, 7L]           73856
              ReLU-8         [-1, 128L, 7L, 7L]               0
         MaxPool2d-9         [-1, 128L, 3L, 3L]               0
           Linear-10                 [-1, 256L]          295168
             ReLU-11                 [-1, 256L]               0
          Dropout-12                 [-1, 256L]               0
           Linear-13                  [-1, 30L]            7710
================================================================
Total params: 396126
Trainable params: 396126
Non-trainable params: 0
----------------------------------------------------------------
None

Training

Run the jupyter notebooks to start the Training.

Training Loss for Single Rectangle Bounding Box

Output for Single Rectangle Bounding Box

The Single Single Rectangle Bounding Box showed and accuracy of 0.99998360718840895

Intermediate output for MultiObject Bounding Box

Single Rectangle Bounding box is completed and the other works are in progress

ToDo

• Single object BBox.
• Save/Load checkpoint.
• Multiple Object BBox withot flip.
• Multiple Object BBox with flipping