swr.cn-south-1.myhuaweicloud.com/ascendhub/mindie:2.3.0-800I-A2-py311-openeuler24.03-lts 

docker run -dit --privileged --ipc=host --name=BlendMask_test  --shm-size=1000g \
    --device=/dev/davinci_manager \
    --device=/dev/devmm_svm \
    --device=/dev/hisi_hdc \
    -v /usr/local/sbin:/usr/local/sbin \
    -v /usr/local/bin/npu-smi:/usr/local/bin/npu-smi \
    -v /usr/local/Ascend/driver:/usr/local/Ascend/driver \
    -v /usr/local/Ascend/firmware:/usr/local/Ascend/firmware \
    -v /usr/local/Ascend/driver/version.info:/usr/local/Ascend/driver/version.info \
    -v /etc/ascend_install.info:/etc/ascend_install.info \
    -v /home:/home \
    -v /data:/data \
    -v /tmp:/tmp \
    97f9fadfa336 \
/bin/bash

docker exec -it BlendMask bash

# 拉取代码仓
git clone https://github.com/aim-uofa/AdelaiDet.git
cd AdelaiDet

# 安装依赖
python setup.py build develop 

# detectron2安装
git clone https://github.com/facebookresearch/detectron2.git 
python -m pip install -e detectron2

# 昇腾适配
vim ./adet/modeling/blendmask/blendmask.py
添加：
import torch_npu
from torch_npu .contrib import transfer_to_npu

https://huggingface.co/ZjuCv/AdelaiDet/blob/main/R_101_3x.pth

# 数据集
https://www.modelscope.cn/datasets/PAI/COCO2017/files

# git-lfs
https://github.com/git-lfs/git-lfs/releases

_BASE_: "Base-BlendMask.yaml"
MODEL:
# 权重路径
  WEIGHTS: "detectron2://ImageNetPretrained/MSRA/R-101.pkl"
  RESNETS:
    DEPTH: 101
SOLVER:
  STEPS: (210000, 250000)
  MAX_ITER: 270000
OUTPUT_DIR: "output/blendmask/R_101_3x"

MODEL:
  META_ARCHITECTURE: "BlendMask"
  MASK_ON: True
  BACKBONE:
    NAME: "build_fcos_resnet_fpn_backbone"
  RESNETS:
    OUT_FEATURES: ["res3", "res4", "res5"]
  FPN:
    IN_FEATURES: ["res3", "res4", "res5"]
  PROPOSAL_GENERATOR:
    NAME: "FCOS"
  BASIS_MODULE:
    LOSS_ON: True
  PANOPTIC_FPN:
    COMBINE:
      ENABLED: False
  FCOS:
    THRESH_WITH_CTR: True
    USE_SCALE: False
DATASETS:
  TRAIN: ("coco_2017_train",)
  TEST: ("coco_2017_val",)
SOLVER:
  IMS_PER_BATCH: 16
  BASE_LR: 0.01  # Note that RetinaNet uses a different default learning rate
  STEPS: (60000, 80000)
  MAX_ITER: 90000
INPUT:
  MIN_SIZE_TRAIN: (640, 672, 704, 736, 768, 800)

cd datasets/
python prepare_thing_sem_from_instance.py

python demo/demo.py     --config-file configs/BlendMask/R_101_3x.yaml     --input /data/AdelaiDet/bus.jpg  --output /data/AdelaiDet/result.jpg   --confidence-threshold 0.35     --opts MODEL.WEIGHTS /data/AdelaiDet/datasets/R_101_3x.pth

# 单卡训练脚本
OMP_NUM_THREADS=1 python tools/train_net.py \
    --config-file configs/BlendMask/R_101_3x.yaml \
    --num-gpus 1 \
OUTPUT_DIR training_dir/R_101_3x_new
# 多卡训练脚本
OMP_NUM_THREADS=1 python tools/train_net.py \
    --config-file configs/BlendMask/R_101_3x.yaml \
    --num-gpus 4 \
OUTPUT_DIR training_dir/R_101_3x_new

pip uninstall -y rapidfuzz
pip install rapidfuzz==2.13.7

if has_gpu:
     devices = defaultdict(list)
     for k in range(torch.cuda.device_count()):
         # cap = "_".join(str(x) for x in torch.cuda.get_device_capability(k))
         dev_cap = torch.cuda.get_device_capability(k)
         cap = "_".join(str(x) for x in (dev_cap if dev_cap is not None else (8, 0)))
         name = torch.cuda.get_device_name(k) + " (" + str(cap) + ")"
         devices[name].append(str(k))
     for name, devs in devices.items():

if has_gpu:
    assert num_gpus_per_machine <= torch.cuda.device_count()
    global_rank = machine_rank * num_gpus_per_machine + local_rank
try:
    dist.init_process_group(
        #backend="NCCL" if has_gpu else "GLOO",
        backend="NCCL",
        init_method=dist_url,
        world_size=world_size,
        rank=global_rank,
        timeout=timeout,