Is there a way to make predictions and visualize the result image by loading a previously trained detectron2 model (without providing any metadata)?

[sushmasuresh28]

I am training a detectron2 model on a custom dataset and saving the "model_final.pth" locally. My question has 2 parts-

1. During testing, I would like to make predictions on any image provided by loading the saved "model_final.pth". Can anyone help me with it? Currently, when I try to load the file and make the predictions, I get no values and the output image with no predictions.

2. Is there a way to print custom class names while making the prediction on a single image (without providing any metadata)?

CASE 1: Currently, by using the below code-
from detectron2.utils.visualizer import ColorMode v = Visualizer(im[:, :, ::-1], MetadataCatalog.get(cfg.DATASETS.TRAIN[0]), scale=1.2) out = v.draw_instance_predictions(outputs["instances"].to("cpu")) cv2_imshow(out.get_image()[:, :, ::-1])
I get -

Although the classes are that of plants, I get them as 'person'

CASE 2: By removing the metadata parameter, I obtain the classes as 'tensor(0)'
from detectron2.utils.visualizer import ColorMode v = Visualizer(im[:, :, ::-1], scale=1.2) out = v.draw_instance_predictions(outputs["instances"].to("cpu")) cv2_imshow(out.get_image()[:, :, ::-1])

Result-

What I would like -

Or the percentages with no class names.

Since I am using a pre-trained model (which I am saving after detectron training in another file), I do not have the metadata registered while making the predictions