wuyang0329 / unet

this is a simple demo for image segmentation.----unet网络进行语义分割的demo,用的数据集是KITTI
97 stars 38 forks source link

Predict output black images #29

Open darvida opened 4 years ago

darvida commented 4 years ago

Hello i have trained the model with 3 classes. The data.py classes looks like: ` BackGround = [0, 0, 0] road = [2, 2, 2]

COLOR_DICT = np.array([BackGround, road])

one = [1, 1, 1]

COLOR_DICT = np.array([BackGround,road,one]) ... self.image_color_mode = "rgb" self.label_color_mode = "rgb"

    self.flag_multi_class = flag_multi_class
    self.num_class = num_classes
    self.target_size = (512, 512)
    self.img_type = 'jpg'

`

I have annotated the images with background color (0,0,0), class 1 (1,1,1) and class 2 (2,2,2). When training it went up to 98% accuracy early in the training, which might indicate that something might be wrong.

Can anyone help me out ?