vi DB/training/learning_rate.py

lr = State(default=0.007) 

in the yaml file
        learning_rate:
            class: DecayLearningRate
            lr: 0.001
            epochs: 1200
	# 学习率要这两个地方一起来成一至后，logs 显示才正常



RuntimeError: transform: failed to synchronize: cudaErrorAssert: device-side assert triggered

I fixed this problem after I change my pytorch version to 1.4.1
    # 解决训练中途出错


see huggingface\DBNet\aliocr_IC15_char_convert.py

# see doc\lang\programming\pytorch\文本检测\DBNET 论文代码都有

字符级标注

测试
CUDA_VISIBLE_DEVICES=0 python demo.py experiments/seg_detector/ic15_resnet18_deform_thre.yaml --image_path datasets/icdar2015/test_images/img_1.jpg --resume /root/final --polygon --box_thresh 0.7 --visualize

CUDA_VISIBLE_DEVICES=0 python eval.py experiments/seg_detector/ic15_resnet18_deform_thre.yaml --resume outputs/workspace/DB/SegDetectorModel-seg_detector/deformable_resnet18/L1BalanceCELoss/model/final --box_thresh 0.55

实测一张图拆出来的每一行一个图，字符级标注：结果完全不得行

todo:
    优化训练原码，自已写