search

Xilinx / Vitis-AI

Vitis AI is Xilinx’s development stack for AI inference on Xilinx hardware platforms, including both edge devices and Alveo cards.
https://www.xilinx.com/ai
Apache License 2.0
1.43k stars 621 forks source link

Yolov5 QAT #1351

Open zijian98 opened 10 months ago

zijian98 commented 10 months ago

Vitis-AI 3.0

Hi, I am trying to perform QAT on yolov5 following this script: https://github.com/Xilinx/Vitis-AI/blob/v3.0/src/vai_quantizer/vai_q_pytorch/example/resnet18_qat.py

However, I encountered this error when trying to export to xmodel:

Traceback (most recent call last):
  File "quant_yolov5_QAT.py", line 30, in <module>
    qat_processor = QatProcessor(model, inputs)
  File "/opt/vitis_ai/conda/envs/vitis-ai-pytorch/lib/python3.7/site-packages/pytorch_nndct/quantization/quant_aware_training.py", line 140, in __init__
    device=self._device)
  File "/opt/vitis_ai/conda/envs/vitis-ai-pytorch/lib/python3.7/site-packages/pytorch_nndct/qproc/base.py", line 182, in __init__
    quantizer.setup(graph, False, lstm_app, custom_quant_ops=custom_quant_ops)
  File "/opt/vitis_ai/conda/envs/vitis-ai-pytorch/lib/python3.7/site-packages/nndct_shared/quantization/base_quantizer.py", line 149, in setup
    custom_quant_ops=custom_quant_ops)
  File "/opt/vitis_ai/conda/envs/vitis-ai-pytorch/lib/python3.7/site-packages/nndct_shared/quantization/quant_info.py", line 45, in __init__
    self.group_graph()
  File "/opt/vitis_ai/conda/envs/vitis-ai-pytorch/lib/python3.7/site-packages/nndct_shared/quantization/quant_info.py", line 81, in group_graph
    quant_groups = self.scan_commander[c](self.Nndctgraph, quant_groups)
  File "/opt/vitis_ai/conda/envs/vitis-ai-pytorch/lib/python3.7/site-packages/nndct_shared/quantization/commander.py", line 76, in SoftFuseReshape
    start_node = quant_groups[p.name][0]
KeyError: 'DetectMultiBackend::DetectMultiBackend/DetectionModel[model]/Detect[model]/Detect[24]/DeQuantStub[dequant_stub]/10910'

Does anyone knows what is this error and how to solve this? Would appreciate it if someone could share some information on how to modify yolo.py during training.

Thank You!

IkrameBeggar commented 8 months ago

Vitis-AI 3.0

Hi, I am trying to perform QAT on yolov5 following this script: https://github.com/Xilinx/Vitis-AI/blob/v3.0/src/vai_quantizer/vai_q_pytorch/example/resnet18_qat.py

However, I encountered this error when trying to export to xmodel:

Traceback (most recent call last):
  File "quant_yolov5_QAT.py", line 30, in <module>
    qat_processor = QatProcessor(model, inputs)
  File "/opt/vitis_ai/conda/envs/vitis-ai-pytorch/lib/python3.7/site-packages/pytorch_nndct/quantization/quant_aware_training.py", line 140, in __init__
    device=self._device)
  File "/opt/vitis_ai/conda/envs/vitis-ai-pytorch/lib/python3.7/site-packages/pytorch_nndct/qproc/base.py", line 182, in __init__
    quantizer.setup(graph, False, lstm_app, custom_quant_ops=custom_quant_ops)
  File "/opt/vitis_ai/conda/envs/vitis-ai-pytorch/lib/python3.7/site-packages/nndct_shared/quantization/base_quantizer.py", line 149, in setup
    custom_quant_ops=custom_quant_ops)
  File "/opt/vitis_ai/conda/envs/vitis-ai-pytorch/lib/python3.7/site-packages/nndct_shared/quantization/quant_info.py", line 45, in __init__
    self.group_graph()
  File "/opt/vitis_ai/conda/envs/vitis-ai-pytorch/lib/python3.7/site-packages/nndct_shared/quantization/quant_info.py", line 81, in group_graph
    quant_groups = self.scan_commander[c](self.Nndctgraph, quant_groups)
  File "/opt/vitis_ai/conda/envs/vitis-ai-pytorch/lib/python3.7/site-packages/nndct_shared/quantization/commander.py", line 76, in SoftFuseReshape
    start_node = quant_groups[p.name][0]
KeyError: 'DetectMultiBackend::DetectMultiBackend/DetectionModel[model]/Detect[model]/Detect[24]/DeQuantStub[dequant_stub]/10910'

Does anyone knows what is this error and how to solve this? Would appreciate it if someone could share some information on how to modify yolo.py during training.

Thank You!

Hello, did you manage to solve the issue please?