Hi, I tested the interface code for predict landmarks by using 300W pretrained weights

fixed ~ you can try the latest version v0.1.6.9 <div class="highlight highlight-so

what is the version of your installed torchlm? check it: <div class="highlight hig

Inference on PyTorch Backend not work about torchlm HOT 5 CLOSED

tongnews commented on May 20, 2024

Inference on PyTorch Backend not work

from torchlm.

Comments (5)

DefTruth commented on May 20, 2024 2

fixed ~ you can try the latest version v0.1.6.9

pip uninstall torchlm 
pip install torchlm>=0.1.6.9

test demo:

    device = "cpu"
    img_path = "./assets/pipnet0.jpg"
    save_path = "./logs/pipnet0_300w.jpg"
    checkpoint = "./pretrained/pipnet/pipnet_resnet101_10x68x32x256_300w.pth"
    image = cv2.imread(img_path)

    torchlm.runtime.bind(faceboxesv2())
    torchlm.runtime.bind(
        pipnet(
            backbone="resnet101",
            pretrained=True,
            num_nb=10,
            num_lms=68,
            net_stride=32,
            input_size=256,
            meanface_type="300w",
            backbone_pretrained=False,
            map_location=device,
            checkpoint=checkpoint
        )
    )
    landmarks, bboxes = torchlm.runtime.forward(image)
    image = torchlm.utils.draw_bboxes(image, bboxes=bboxes)
    image = torchlm.utils.draw_landmarks(image, landmarks=landmarks)

    cv2.imwrite(save_path, image)

you can set backbone_pretrained as False, because you already have a pretrained file in your local device, thus it's not necessarily to download the backbone weight from internet.

from torchlm.

DefTruth commented on May 20, 2024

what is the version of your installed torchlm? check it: