This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

TexasInstruments / edgeai-modelzoo 例程 问题

您好,我在使用咱们公司 Github官网的例程进行训练 :https://github.com/TexasInstruments/edgeai-modelzoo/tree/main/models/vision/segmentation

我训练 cityspace 官方数据集,可以进行训练,并推理。

我将自己的数据集修改为 cityspaces 格式进行训练,可以训练,但是推理出来的结果不正确,类似于马赛克的效果。

以下是我的 cityscapes_plus.py 参数设置:

"""CityscapesLoader: Data is derived from CityScapes, and can be downloaded from here: https://www.cityscapes-dataset.com/downloads/
    Many Thanks to @fvisin for the loader repo: """>github.com/.../cityscapes.py"""

    colors = [[128, 64, 128], [244, 35, 232], [0, 0, 0]]
    label_colours = dict(zip(range(2), colors))
    ignore_index = 255
    void_classes = [-1, ignore_index]
    valid_classes = [0, 1]
    # class_names = ['road', 'pudele']

    class_map = dict(zip(valid_classes, range(len(valid_classes))))
    num_classes_ = len(valid_classes)

    class_weights_ = np.array([0.22567085, 1.89944273], dtype=float)
我只有两个类:road(0)和puddle(1),其余为255。
问题:
1、请问,是我的参数设置错了吗?还是其他位置的赛马需要修改呀。
2、对于我自己的数据,采用的是 cityspace 官方转换代码转的,是数据转换格式的问题吗?