我想在YOLO上训练模型，但在第一步就出现了nan和-nan。我有300张不同尺寸的图像（几乎都是600*600），并且只有一个类要检测。在之前使用100张图像时，检测准确率达到了75%，但我希望能得到更好的结果。

tiny_yolo.cfg

[net]batch=64subdivisions=8width=416height=416channels=3momentum=0.9decay=0.0005angle=0saturation = 1.5exposure = 1.5hue=.1learning_rate=0.001max_batches = 120000policy=stepssteps=-1,100,80000,100000scales=.1,10,.1,.1[convolutional]batch_normalize=1filters=30size=3stride=1pad=1activation=leaky[maxpool]size=2stride=2[convolutional]batch_normalize=1filters=30size=3stride=1pad=1activation=leaky[maxpool]size=2stride=2[convolutional]batch_normalize=1filters=30size=3stride=1pad=1activation=leaky[maxpool]size=2stride=2[convolutional]batch_normalize=1filters=30size=3stride=1pad=1activation=leaky[maxpool]size=2stride=2[convolutional]batch_normalize=1filters=30size=3stride=1pad=1activation=leaky[maxpool]size=2stride=2[convolutional]batch_normalize=1filters=30size=3stride=1pad=1activation=leaky[maxpool]size=2stride=1[convolutional]batch_normalize=1filters=30size=3stride=1pad=1activation=leaky###########[convolutional]batch_normalize=1size=3stride=1pad=1filters=30activation=leaky[convolutional]size=1stride=1pad=1filters=30activation=linear[region]anchors = 0.738768,0.874946,  2.42204,2.65704,  4.30971,7.04493,  10.246,4.59428,  12.6868,11.8741bias_match=1classes=1coords=4num=5softmax=1jitter=.2rescore=1small_object=1object_scale=5noobject_scale=1class_scale=1coord_scale=1absolute=1thresh = .6random=1

我按照80-20的比例分割了训练和测试数据，我使用的是这个darknet

请帮帮我！

回答：

我忘了配置darknet的Makefile。我使用的是Google Colab，首先必须定义使用GPU和CUDNN。