Revision history [back]

the 1st convolution layer in your network complains, that the number of channnels did not fit. ;)

we don't know, what kind of model you're using, but most likely it expects BGR or RGB input, not grayscale, so try a

Imgproc.cvtColor(frame, frame, Imgproc.COLOR_RGBA2BGR);

instead.

the 1st convolution layer in your network complains, that the number of channnels channels in your input did not fit. ;)

we don't know, what kind of model you're using, but most likely it expects BGR or RGB input, not grayscale, so try a

Imgproc.cvtColor(frame, frame, Imgproc.COLOR_RGBA2BGR);

instead. instead.

the 1st convolution layer in your network complains, that the number of channels in your input did not fit. ;)

we don't know, what kind of model you're using, but most likely it expects BGR or RGB input, not grayscale, so try a

Imgproc.cvtColor(frame, frame, Imgproc.COLOR_RGBA2BGR);

instead.

the 1st convolution layer in your network complains, that the number of channels in your input did not fit. ;)

we don't know, what kind of model you're using, but most likely it expects BGR or RGB input, not grayscale, so try a

Imgproc.cvtColor(frame, frame, Imgproc.COLOR_RGBA2BGR);

instead.

edit:

(see 1st comment !)

you probably took the code from the android sample here, but mistook someting:

leNet is a classification model, NOT a detection one. you simply cannot infer bounding boxes from it.

all you can do is apply minMaxLoc() on the softmax output, and infer the classID from there.there:

Mat detections = net.forward("Softmax").reshape(1,1);
MinMaxLocResult mm = Core.minMaxLoc(detections);
int classID = mm.maxPos.x;

the 1st convolution layer in your network complains, that the number of channels in your input did not fit. ;)

we don't know, what kind of model you're using, but most likely it expects BGR or RGB input, not grayscale, so try a

Imgproc.cvtColor(frame, frame, Imgproc.COLOR_RGBA2BGR);

instead.

edit:

(see 1st comment !)

you probably took the code from the android sample here, but mistook someting:something:

leNet is a classification model, NOT a detection one. you simply cannot infer bounding boxes from it.

all you can do is apply minMaxLoc() on the softmax output, and infer the classID from there:

Mat detections = net.forward("Softmax").reshape(1,1);
MinMaxLocResult mm = Core.minMaxLoc(detections);
int classID = mm.maxPos.x;

the 1st convolution layer in your network complains, that the number of channels in your input did not fit. ;)

we don't know, what kind of model you're using, but most likely it expects BGR or RGB input, not grayscale, so try a

Imgproc.cvtColor(frame, frame, Imgproc.COLOR_RGBA2BGR);

instead.

edit:

(see 1st comment !)

you probably took the code from the android sample here, but mistook something:

leNet is a classification model, NOT a detection one. you simply cannot infer bounding boxes from it.

all you can do is apply minMaxLoc() on the softmax output, and infer the classID from there:

Mat detections = net.forward("Softmax").reshape(1,1);
MinMaxLocResult mm = Core.minMaxLoc(detections);
int classID = mm.maxPos.x;
mm.maxLoc.x;

the 1st convolution layer in your network complains, that the number of channels in your input did not fit. ;)

we don't know, what kind of model you're using, but most likely it expects BGR or RGB input, not grayscale, so try a

Imgproc.cvtColor(frame, frame, Imgproc.COLOR_RGBA2BGR);

instead.

edit:

(see 1st comment !)

you probably took the code from the android sample here, but mistook something:

leNet is a classification model, NOT a detection one. you simply cannot infer bounding boxes from it.

all you can do is apply minMaxLoc() on the softmax output, and infer the classID from there:

Mat detections = net.forward("Softmax").reshape(1,1);
MinMaxLocResult mm = Core.minMaxLoc(detections);
int classID = mm.maxLoc.x;
double probability = mm.maxVal;