I'm trying to apply a retrained model of mobilenet_v2 presented in https://github.com/balajisrinivas/Face-Mask-Detection

The main part of my adapted code is the following:

faces = []
locs = []
preds = []

for coord in coords:

    x = person[0]
    y = person[1]
    w = coord [2]
    h = person[3]

    # extract the face ROI, convert it from BGR to RGB channel
    # ordering, resize it to 224x224, and preprocess it
    face = frame[y:y+h, x:x+h]
    face = cv2.cvtColor(face, cv2.COLOR_BGR2RGB)

    if (face.shape) > (224, 224):
        face = cv.resize(face, (224, 224))
    face = img_to_array(face)
    face = preprocess_input(face) #from tensorflow.keras.applications.mobilenet_v2

    # add the face and bounding boxes to their respective lists
    locs.append((x, y, w, h))

    # only make a predictions if at least one face was detected
    if len(faces) > 0:
        # for faster inference we'll make batch predictions on *all*
        # faces at the same time rather than one-by-one predictions
        faces = np.array(faces, dtype="float32")
        preds = maskNet.predict(faces, batch_size=32) #maskNet = load_model("mask_detector.model")

The error is raised when I try to apply maskNet.predict, which is the is the mobilenet_v2 model saved in https://github.com/balajisrinivas/Face-Mask-Detection/blob/master/mask_detector.model.

The error message is the following: ValueError: Negative dimension size caused by subtracting 3 from 1 for '{{node model/Conv1/Conv2D}} = Conv2D[T=DT_FLOAT, data_format="NHWC", dilations=[1, 1, 1, 1], explicit_paddings=[], padding="VALID", strides=[1, 2, 2, 1], use_cudnn_on_gpu=true](model/Conv1_pad/Pad, model/Conv1/Conv2D/ReadVariableOp)' with input shapes: [?,1,1,3], [3,3,3,32].


I found out my mistake!

I shouldn't use a condition before resizing the frame:

face = cv.resize(face, (224, 224)) #without if (face.shape) > (224, 224):

I've used the condition because I had to do that for SSD Resnet in the past, but I just realized it's not necessary for mobilenet_v2


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.