The test with the landmark.py you applied is not accurate.

[XiaXuehai]

trafficstars

I use the landmark and the caffemodel you applied to test the image,and the result wasn't accurate. and I see the classification.cpp caffe applied.The input testing image should minus mean_file.Do you think it's a factor of inaccuracy.

[luoyetx]

The code here process the image. The model here is still not comparable to the original paper.

[XiaXuehai]

THX. But the mean_file in caffe's example is the mean of the training image.And it's not the same as the following:

def _processImage(self, imgs):
        imgs = imgs.astype(np.float32)
        for i, img in enumerate(imgs):
            m = img.mean()
            s = img.std()
            imgs[i] = (img - m) / s
        return imgs

Does it affect the result?

[luoyetx]

@XiaXuehai Yes. Caffe support an external mean_file for data layer. But I didn't use it. Actually, I did the image processing in my own code and give a raw data input to the network.

[hugowangyj]

Hello, luoyetx, what's your lowest loss on the test dataset? I find mine is a large gap with the paper.

[luoyetx]

@hugowangyj You can find some data in this pdf.

[hugowangyj]

Yes, but it is still a large lag behind the original paper. Have you the latest results?