Created by: TianyouChen

我想用剪枝出来的模型进行量化，发现参数加载不匹配，网络加载原始yolov3，但是参数加载剪枝后的，这个该怎么配置修改让参数匹配并能量化训练？ 详细信息如下： 训练指令：

python3 slim/quantization/train.py --not_quant_pattern slim_output \
 --eval --c ./configs/yolov3_mobilenet_v1_Headshoulder.yml -o max_iters=30000 \
save_dir=./output/prune_quant_yolov3 LearningRate.base_lr=0.0001 LearningRate.schedulers="[!PiecewiseDecay {gamma: 0.1, milestones: [10000]}]" pretrain_weights=output/prue_Headshoulder/yolov3_mobilenet_v1_Headshoulder/best_model

报错信息：

2020-03-30 09:53:24,269-INFO: 1690 samples in file dataset/voc/test.txt
2020-03-30 09:53:24,270-INFO: places would be ommited when DataLoader is not iterable
W0330 09:53:25.296880  6945 device_context.cc:237] Please NOTE: device: 0, CUDA Capability: 70, Driver API Version: 10.1, Runtime API Version: 9.0
W0330 09:53:25.300591  6945 device_context.cc:245] device: 0, cuDNN Version: 7.6.
2020-03-30 09:53:28,249-INFO: Loading parameters from output/prue_Headshoulder/yolov3_mobilenet_v1_Headshoulder/best_model...
Traceback (most recent call last):
  File "slim/quantization/train.py", line 314, in <module>
    main()
  File "slim/quantization/train.py", line 193, in main
    ignore_params=ignore_params)
  File "/home/chenchaocun/PaddleDetection_3.27/ppdet/utils/checkpoint.py", line 142, in load_params
    fluid.io.set_program_state(prog, state)
  File "/usr/local/lib/python3.5/dist-packages/paddle/fluid/io.py", line 1942, in set_program_state
    .format(orig_para_np.shape, para.name, new_para_np.shape)
AssertionError: Shape not matching: the Program requires a parameter with a shape of ((512, 1024, 1, 1)), while the loaded parameter (namely [ yolo_block.0.0.0.conv.weights ]) has a shape of  ((51, 1024, 1, 1)).

多谢指教～