How to use other versión of YOLO

Hi, this time I want to ask How difficult is to change the version YOLO that is used in the deep perception node of the robot.
¿I have to create a deep perception node or I only need to do some changes in the node to use another version of YOLO?

Hi @Humberto_Andres_Hida, I missed this thread before! Sorry about that! If you’re still interested in using other versions of YOLO with Stretch Deep Perception, I do recommend taking a look at the API defined in the ObjectDetector class, which is currently calling YOLOv3. Implementing the apply_to_image() method is all you would need to swap in a different model for 3d detections.


I hope this is not too unrelated: I want to train a custom model by adding images to the existing COCO dataset to detect specific objects related to my project.

I plan to build the model according to this tutorial.

Is there any specification I need to keep in mind when replacing the old Yolo Model on the Stretch?

Hey @roboor, yup, implementing the same apply_to_image() by extending the ObjectDetection class will work for you as well. You would swap the YOLO checkpoint file with the one that you trained.

1 Like

Hey @bshah,

Thanks for your reply.

Can we also swap the YOLO checkpoint file not only with newer versions of YOLO but with other models? I want to train my data with the TensorFlow ssd-mobilenet-v2-fpnlite-320 for example.

Hey @roboor, every model will have different ways to perform the inference and retrieve the detections, so you would need to implement the apply_to_image() method for the specific model you’re using. I can help you with it if you start a new thread and include the checkpoint file and your code for inferencing.