Can you share me pretrain model on other link such that google driver?

[ThorPham]

I'm not chinese . I can't down load pretrain model on baidu.

[universea]

I will try that later.

[ThorPham]

Thank for your help .

[universea]

Sorry, In our country we can't use google service .Can you give me your e-mail and I send the params to you ?

[ThorPham]

@universea my email [email protected] . Thank you so much

[ThorPham]

@universea how to visualize output same as demo. I try predict model and output is binary image. How to i merger these image to image color with segmentation.

[universea]

you can replace this in predict.py and use these color images to make gif file by other image tools.

def saveImage(img,path):
label_rgb = np.zeros((img.shape[0], img.shape[1], 3), dtype=np.uint8)
label_rgb[img == 0] = [ 0, 0, 0]
label_rgb[img == 1] = [ 70, 130, 180]
label_rgb[img == 2] = [119, 11, 32]
label_rgb[img == 3] = [220, 220, 0]
label_rgb[img == 4] = [102, 102, 156]
label_rgb[img == 5] = [128, 64, 128]
label_rgb[img == 6] = [190, 153, 153]
label_rgb[img == 7] = [128, 128, 0]
label_rgb[img == 8] = [255, 128, 0]
img = cv2.resize(label_rgb, (4000,4000), interpolation=cv2.INTER_NEAREST)
img = cv2.warpPerspective(img, Minv, (3384, 1710),flags=cv2.INTER_NEAREST)
cv2.imwrite(path, img)

[ThorPham]

your output very bad . Have something wrong ?
https://github.com/ThorPham/lane_segmetation

[universea]

Maybe the params was bad , I send you the new params and some infer images.you should rename the params as "unet.params"

[ThorPham]

Thank you very much for your support. But the output still very bad. I think something wrong in file predict.py . Can you share me your file predict and some image your test.

[ThorPham]

I see proplem why output my predict size 1024x1024 but your output is 3384x1710