Re-trained SSD Model to detect kittens!

Code can be found on Github. Or a side-by-side comparison of the networks here.

I re-trained the SSD-Mobilenet-v1 checkpoint provided by jetson-inference with a set of 2500 images that I labeled with labelImg.

These kittens were living with me for a few weeks, so I had no choice but to train a model to tell them apart! I enjoyed the experience of training my first model using my own data to provide inference for a task in my real world.

I appreciate any tips or comments on my work, as I am just getting started in this awesome field!

1 Like


This is so cute !!!
Thanks for sharing this with us.

Here is advice for you.
It seems that the accuracy slightly drops when the occlusion occurs.
Maybe you can tune/design the non-maximum suppression based on your use case to improve.


1 Like