Running Inference on the Custom Dataset

[Brian96086]

Hi,
Thanks for the interesting and impressive work! I would like to run inference on a custom dataset. In the docs, running inference is involved with running train.py, which is strongly tied with the training setup. Do you have any suggestions or scripts so that I can run inference effectively and independently of training?

Thanks again for your time and effort !

[xk-huang]

Hi @Brian96086,

There are two solutions to inference the models on the custom dataset.

1. The API of SCA model:

   segment-caption-anything/scripts/apps/sca_app.py

   Line 158 in ceaca87

   def run(args, input_image, input_points=None, input_boxes=None):

   
   Also see: https://github.com/xk-huang/segment-caption-anything/blob/main/scripts/notebooks/auto_generator.ipynb.

2. Prepare the dataset into VG/COCO format, and use the scripts in https://github.com/xk-huang/benchmark-referring-vllm.

I would prefer the first one for the maximum flexibility.

[Brian96086]

Hi @xk-huang ,
Thanks for the clarification and instruction. Really appreciate your help!