dbnet.nntc.config.yaml

---
name: dbnet
model: $(home)/inference
dataset_path: $(home)/data/icdar2015
output_path: $(root)/output/$(name)/inference
time: true
precision: true
gops: [5.255]

div_info: div=0.229:0.224:0.225
mean_info: mean_value=0.485:0.456:0.406
size_info: resize_h=736,resize_w=1280;scale=0.00392156862
fp_batch_sizes: [1]

fp_loops:
  - build_env:
      - BMCOMPILER_LAYER_DTYPE_MODE=
    fp_outdir_template: "{}b.fp32.compilation"

BM1684X:
  fp_compile_options:
    python3 -m bmpaddle
        --model=$(model)
        --shapes [4,3,736,1280]
        --net_name $(name)
        --target BM1684X
        --cmp=False
        --enable_profile True

  #  time_only_cali:
  #    python3 -u -m ufw.cali.cali_model
  #      --net_name $(name)
  #      --model $(model)/inference.pdmodel
  #      --input_shapes [1,3,736,1280]
  #    --cali_image_preprocess '$(size_info),$(mean_info),$(div_info),bgr2rgb=1'
  #      --cali_image_path $(dataset_path)/ch4_quanti_data
  #      --debug_cmd='to_umodel_not_cmp;quant_1684X;not_suspend;not_call_bmnetu'
  #      --target=BM1684X
  #      --cali_iterations 1
  #      --test_iterations 1
  #      --input_names "x"
  #      --output_names "sigmoid_0.tmp_0"

  #  bmnetu_batch_sizes: [1]

  #  int8_model: $(workdir)/$(name)_bmpaddle_deploy_int8_unique_top.prototxt
  #  int8_weight: $(workdir)/$(name)_bmpaddle.int8umodel

  #  bmnetu_options:
  #    --model $(int8_model)
  #    --weight $(int8_weight)
  #    --net_name $(name)
  #    --target BM1684X
  #    --cmp=False
  #    --v=4
  #    --opt 2
  #    --output_as_fp32 sigmoid_0.tmp_0
  #    --seed=42
  #    --enable_profile=True

  build_env:
    - BMCOMPILER_EXPORT_GRAPH=1