Created by: TianyouChen

最新版本paddleDetection训练自己类别的时候出错，而旧版本是可以换成自己的标签数据训练的

背景：最近用paddleDetecion做量化，而旧版版本有些api不支持，就把paddle、paddleslim、paddleDetection全部换成最新版，最后成功跑起VOC 21类目标检测检测量化训练，但是当我改标签跑自己数据时候出错了： 指令： python3 slim/quantization/train.py --not_quant_pattern slim_output --eval --c ./configs/ssd/ssd_mobilenet_v1_Headshoulder.yml -o max_iters=30000 save_dir=./output/mobilenetv1_slim LearningRate.base_lr=0.0001 LearningRate.schedulers="[!PiecewiseDecay {gamma: 0.1, milestones: [10000]}]" pretrain_weights=/home/chenchaocun/PaddleDetection_slim/best_model 错误信息： pretrain_weights: /home/chenchaocun/PaddleDetection_slim/best_model save_dir: ./output/mobilenetv1_slim snapshot_iter: 2000 test_feed: SSDTestFeed train_feed: SSDTrainFeed use_gpu: true weights: output/ssd_mobilenet_v1_Headshoulder_slim/model_final/

Traceback (most recent call last): File "slim/quantization/train.py", line 314, in main() File "slim/quantization/train.py", line 111, in main inputs_def = cfg['TrainReader']['inputs_def'] KeyError: 'TrainReader'

然后我不做量化训练，直接训练自己数据，发现只要我把PaddleDetection_slim/tools/ppdet/data/source/voc.py这里的标签改成自己的数据标签加载自己的数据，训练就出现错误： pretrain_weights: https://paddlemodels.bj.bcebos.com/object_detection/ssd_mobilenet_v1_coco_pretrained.tar save_dir: output snapshot_iter: 2000 test_feed: SSDTestFeed train_feed: SSDTrainFeed use_gpu: true weights: output/ssd_mobilenet_v1_Headshoulder_slim/model_final/

Traceback (most recent call last): File "tools/train.py", line 323, in main() File "tools/train.py", line 114, in main inputs_def = cfg['TrainReader']['inputs_def'] KeyError: 'TrainReader'

而之前旧版本的PaddleDetection是可以训练自己数据的，请问下这个怎么解决？