0

I trained a model, now I would like to use it to detect objects in images. Using the DefaultDetector only the boundyboxes are returned, I would need the masks. I saw that you can also perform inference with this method:

model.eval()
with torch.no_grad():
    outputs = model(inputs)

I think that's what he should use. The problem is that I don't know how to set the inputs, starting with images.

import torch
import glob
cfg = get_cfg()

cfg.merge_from_file(model_zoo.get_config_file("COCO-InstanceSegmentation/"
                                              "mask_rcnn_R_101_FPN_3x.yaml"))
cfg.MODEL.WEIGHTS = os.path.join(cfg.OUTPUT_DIR, "model_final.pth")
cfg.SOLVER.IMS_PER_BATCH = 1
cfg.MODEL.ROI_HEADS.NUM_CLASSES = 1 # only has one class
cfg.INPUT.FORMAT = "BGR"
#Just run these lines if you have the trained model im memory
cfg.MODEL.ROI_HEADS.SCORE_THRESH_TEST = 0.7 # set the testing threshold for this model
#build model
model = build_model(cfg)

DetectionCheckpointer(model).load("output/model_final.pth")

model.eval()#make sure its in eval mode

image = cv2.imread("/kaggle/working/detectron2/images/73-ab1.jpg")
height, width = image.shape[:2]
image = torch.as_tensor(image.astype("float32").transpose(2, 0, 1))
image = ImageList.from_tensors([image])

with torch.no_grad():
    inputs = image
    outputs = model(inputs)

Unfortunately, however, I think I'm wrong, can someone enlighten me?

martineau
  • 119,623
  • 25
  • 170
  • 301
Claudio Russo
  • 106
  • 1
  • 7

1 Answers1

0

See the Model Input Format for the builtin models.

Basically, the model in your code is not expecting an ImageList object, but a list of dicts where each dict needs to provide specific information about one image, as explained in the documentation linked above.

So, your inference code needs to be corrected to the following.

image = cv2.imread("/kaggle/working/detectron2/images/73-ab1.jpg")
height, width = image.shape[:2]
image = torch.as_tensor(image.astype("float32").transpose(2, 0, 1))
inputs = [{"image": image, "height": height, "width": width}]

with torch.no_grad():
    outputs = model(inputs)

You can also see this in the code - the forward method of the GeneralizedRCNN class.

zepman
  • 671
  • 6
  • 14