feature/add torchserve detectron2 #3355
Conversation
|
|
||
| --- | ||
|
|
||
| ## Contributors |
There was a problem hiding this comment.
Please remove this section
|
|
||
| --- | ||
|
|
||
| ## Usage |
There was a problem hiding this comment.
Please take a look at other examples and show all the steps in this README. Anyone should be able to replicate the example looking at just the README.
There was a problem hiding this comment.
Thank you for the guidance! I’ve now updated the README file.
| logger.info(f"Inference started for a batch of {len(model_input)}.") | ||
|
|
||
| outputs = [] | ||
| for idx, image in enumerate(model_input): |
There was a problem hiding this comment.
Can detectron2 process a batch of images? Can we send the batch instead of looping over each image
There was a problem hiding this comment.
Yes, Detectron2 can process a batch of images, and we can send them.
| List[str]: List of predictions. | ||
| """ | ||
| logger.info("Handling request...") | ||
| start_time = time.time() |
There was a problem hiding this comment.
We already have a utility to measure the time, please check this handler
|
HI @Mudassar-MLE Thanks for the PR. Can you please add some context in the PR and write a short description |
…dassar-MLE/serve into feature/torchserve-detectron2
Add Detectron2 Support to TorchServe Object Detection Examples (#3344)
This pull request introduces support for Detectron2 models in TorchServe, addressing issue #3344. It includes a custom handler designed to ensure seamless deployment on both CPU and GPU environments, automatically managing device compatibility. Additionally, the example provides a requirements.txt file for dependencies and a detailed README to guide users in deploying pre-trained models from the Detectron2 Model Zoo as well as custom fine-tuned models. The updates are placed under examples/object_detector/detectron2 as suggested by the maintainers.
The contribution aims to simplify the deployment process for Detectron2 models in TorchServe, enhancing its usability for developers and making it a more commercially viable option for object detection tasks.