Can you please let me know if you know of any examples of object detection models such as Faster r cnn that output object attributes (color, etc.) in conjunction with object labels and box location information?
Hello,
I know the answer will sound generic, but if you have any kind of training dataset, so data is annotated with labels of your need. You might try to use any of RCNN models and just add a new head, which will be trained with the rest of the model frozen. Just use the backbone as the source for the features and build you topology on top of it.
If you provide any additional info, based on the purpose of your model, and if you have the training datasets, I would be happy to follow up more.
Edit:
I was looking for exact page in documentation:
Feature extraction for model inspection — Torchvision main documentation (pytorch.org)
Thank you for the reply.
Do you mean “new head” is a unit in a network like RPN Head or RoI Head？
The model I am tring to build is a model that detects plants from image and outputs the plant name and leaf age (degree of plant growth). The dataset is in PASCAL VOC format, with each plant box, label and leaf age (numerical value) recorded.
Best regards.
Exactly like this. Nothing stops you from creating any kind of new head :).
Base the features on the backbone, and fine-tune the network for the problem.
Thanks for the advice.
I will try again and come back here if I find something I don’t understand.
Thank you.