search

NVIDIA / DALI

A GPU-accelerated library containing highly optimized building blocks and an execution engine for data processing to accelerate deep learning training and inference applications.
https://docs.nvidia.com/deeplearning/dali/user-guide/docs/index.html
Apache License 2.0
5.15k stars 622 forks source link

KITTI Data Format Support #2942

Open adithya-p opened 3 years ago

adithya-p commented 3 years ago

I see that it supports coco format data. I want to know if DALI supports KITTI format data? Please let me know if DALI supports KITTI format transforming images and the bounding boxes (primarily xmin, ymin, xmax, ymax). Any leads would be appreciated.

consider an example,

images/1.png annotations/1.txt

car 0 0 0 522 291 562 320 0 0 0 0 0 0 0
bus 0 0 0 811 262 919 298 0 0 0 0 0 0 0

I want to rotate the image and the corresponding bounding boxes for object detection

JanuszL commented 3 years ago

Hi @adithya-p , DALI doesn't have a dedicated reader for the KITTI data set format. You can write one in python using the external source operator as in this example . You can play with parallel and prefetch_queue_depth to obtain the best performance.

adithya-p commented 3 years ago

I see these issues #2854 #2900 kind of overlap with the annotations format (xmin, ymin, xmax, ymax) but I don't see how to load the annotations and generate modified annotations w.r.t the transformed image?

JanuszL commented 3 years ago

@adithya-p I suggest starting from the mentioned external_source example, writhe a python code that loads the data and passes it to the DALI pipeline.