I am at an intermediate level in Deep Learning. I wanted to implement object detection from scratch. Is the scratch implementation possible by a single person or will i have to use an high level API to do so. I am confused and don't know where to start.

  • $\begingroup$ From scratch, you mean using a deep learning package such as PyTorch / Keras / TensorFlow or do you want to write the entire math part of it as well? $\endgroup$
    – Nischal Hp
    Feb 17, 2020 at 16:22
  • $\begingroup$ yes i want to write it with tensorflow or pytorch with the already known algos $\endgroup$
    – KJS
    Feb 23, 2020 at 11:58
  • $\begingroup$ sry for the late reply $\endgroup$
    – KJS
    Feb 23, 2020 at 12:00

1 Answer 1


As you are looking to implement already known networks with a framework such as pytorch/tensorflow, I would definitely suggest reading into this post : https://towardsdatascience.com/object-detection-and-tracking-in-pytorch-b3cf1a696a98

Given that you wanted to implement this from scratch, its important you understand what the following:

  1. Bounding boxes in images
  2. Loss function - Intersection over union
  3. Network architecture - You can use nettron to learn about the network architecture once you have written it.
  4. Visualize the learning of your model after every 'x' epochs, to gain an intuituve understanding of what your network is trying to learn given the training dataset.
  • $\begingroup$ Thankyou I feel a sense of approach now $\endgroup$
    – KJS
    Feb 28, 2020 at 14:48
  • $\begingroup$ Cool, you can mark the question answered in case you got the answer you were looking for. Helps close the question. $\endgroup$
    – Nischal Hp
    Feb 28, 2020 at 19:12

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.