DIGITS API documentation for object detection (using DetectNet) using Caffe

AI & Data Science Deep Learning (Training & Inference) DIGITS (Locked)
1

Hi all,

I’m looking for REST API documentation to train an object detection model using my own dataset (which has labels in KITTI data format) using DetectNet with Caffe framework. I looked into REST API documentation for classification and regression tasks in the following link -[url]DIGITS/API.md at master · NVIDIA/DIGITS · GitHub

But, this link doesn’t have good explanation on how to use a custom object detection dataset (with labels in KITTI format) to train DetectNet model using Caffe. Essentially, I’m looking for an API call which I could use to create a training job on DIGITS for object detection and to obtain model file (snapshot_iter_xxx.caffemodel) after training is complete. Please point me in the right direction. Any help would be appreciated.

2

Hi,

Since this is a pretty advanced use case, I encourage you to take a look at create_model classmethod in digits/model/images/generic/test_views.py. It shows how to construct the complete payload of POST method to /models/images/generic.