Conversion of Owl-ViT model to ".engine" file which is used in metropolis GenAI application

Accelerated Computing Intelligent Video Analytics Metropolis Microservices for Jetson
1

Hello all,

I am using mmj_genai application. It is using the google’s Owl-ViT model.

I want to use mmj_genai application to detect custom objects. So first I want to train the Owl-ViT model and then convert it to “.engine” file which is needed for mmj_genai application. I want to know how can I convert the Owl-ViT model to .engine file? In the GitHub they are using checkpoint only. How can I save the model from checkpoint and then convert it to “.engine” file?

2

There is guide for convert model to engine file (Build the TensorRT engine for the OWL-ViT vision encoder) in:GitHub - NVIDIA-AI-IOT/nanoowl at cfef75a8ad5fb8be0e3beb501a763661a9336d1d.
The model we use in the mmj_genai example is based on OWL-ViT from google. The OWL-ViT github repository has some resources on how to train it. scenic/scenic/projects/owl_vit at main · google-research/scenic · GitHub

1 Like
3

I looked at the Nanoowl github. Somehow I missed this. Thanks.

The training steps on scenic github repo is not working though. It is giving error related to DECODER for the dataset (‘lvis’ or ‘coco’ datasets).

4

Have you ever tried fine-tuning the model?

5

Please file new topic if you have more question. Thanks!

6

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.