update mmdet

Author: yzbx

det-mmdetection-tmi/mining-template.yaml

@@ -1,3 +1,3 @@
 shm_size: '128G'
-mining_algorithm: cald
+mining_algorithm: aldd
 class_distribution_scores: '' # 1.0,1.0,0.1,0.2

det-mmdetection-tmi/start.py

@@ -53,7 +53,15 @@ def _run_mining(cfg: edict) -> None:
 
 
 def _run_infer() -> None:
-    command = 'python3 ymir_infer.py'
+    gpu_id: str = str(cfg.param.get('gpu_id', '0'))
+    gpu_count = len(gpu_id.split(','))
+
+    if gpu_count <= 1:
+        command = 'python3 ymir_infer.py'
+    else:
+        port = find_free_port()
+        command = f'python3 -m torch.distributed.launch --nproc_per_node {gpu_count} --master_port {port} ymir_infer.py'  # noqa
+
     logging.info(f'start infer: {command}')
     subprocess.run(command.split(), check=True)
     logging.info("infer finished")

det-mmdetection-tmi/ymir_infer.py

@@ -1,20 +1,27 @@
 import argparse
+import os
 import os.path as osp
 import sys
 import warnings
 from typing import Any, List
 
 import cv2
 import numpy as np
+import torch.distributed as dist
 from easydict import EasyDict as edict
 from mmcv import DictAction
-from mmdet.apis import inference_detector, init_detector
-from mmdet.utils.util_ymir import get_best_weight_file
+from mmcv.runner import init_dist
 from tqdm import tqdm
-from ymir_exc import dataset_reader as dr
-from ymir_exc import env
 from ymir_exc import result_writer as rw
-from ymir_exc.util import YmirStage, get_merged_config, write_ymir_monitor_process
+from ymir_exc.util import (YmirStage, get_merged_config, write_ymir_monitor_process)
+
+from mmdet.apis import inference_detector, init_detector
+from mmdet.apis.test import collect_results_gpu
+from mmdet.utils.util_ymir import get_best_weight_file
+
+LOCAL_RANK = int(os.getenv('LOCAL_RANK', -1))  # https://pytorch.org/docs/stable/elastic/run.html
+RANK = int(os.getenv('RANK', -1))
+WORLD_SIZE = int(os.getenv('WORLD_SIZE', 1))
 
 
 def parse_option(cfg_options: str) -> dict:
@@ -80,8 +87,9 @@ def __init__(self, cfg: edict):
         cfg_options = parse_option(options) if options else None
 
         # current infer can only use one gpu!!!
-        gpu_ids = cfg.param.get('gpu_id', '0')
-        gpu_id = gpu_ids.split(',')[0]
+        # gpu_ids = cfg.param.get('gpu_id', '0')
+        # gpu_id = gpu_ids.split(',')[0]
+        gpu_id = max(0, RANK)
         # build the model from a config file and a checkpoint file
         self.model = init_detector(config_file, checkpoint_file, device=f'cuda:{gpu_id}', cfg_options=cfg_options)
 
@@ -90,26 +98,47 @@ def infer(self, img):
 
 
 def main():
+    if LOCAL_RANK != -1:
+        init_dist(launcher='pytorch', backend="nccl" if dist.is_nccl_available() else "gloo")
+
     cfg = get_merged_config()
 
-    N = dr.items_count(env.DatasetType.CANDIDATE)
+    with open(cfg.ymir.input.candidate_index_file, 'r') as f:
+        images = [line.strip() for line in f.readlines()]
+
+    max_barrier_times = len(images) // WORLD_SIZE
+    if RANK == -1:
+        N = len(images)
+        tbar = tqdm(images)
+    else:
+        images_rank = images[RANK::WORLD_SIZE]
+        N = len(images_rank)
+        if RANK == 0:
+            tbar = tqdm(images_rank)
+        else:
+            tbar = images_rank
     infer_result = dict()
     model = YmirModel(cfg)
-    idx = -1
 
     # write infer result
     monitor_gap = max(1, N // 100)
     conf_threshold = float(cfg.param.conf_threshold)
-    for asset_path, _ in tqdm(dr.item_paths(dataset_type=env.DatasetType.CANDIDATE)):
+    for idx, asset_path in enumerate(tbar):
         img = cv2.imread(asset_path)
         result = model.infer(img)
         raw_anns = mmdet_result_to_ymir(result, cfg.param.class_names)
 
+        # batch-level sync, avoid 30min time-out error
+        if WORLD_SIZE > 1 and idx < max_barrier_times:
+            dist.barrier()
+
         infer_result[asset_path] = [ann for ann in raw_anns if ann.score >= conf_threshold]
-        idx += 1
 
         if idx % monitor_gap == 0:
-            write_ymir_monitor_process(cfg, task='infer', naive_stage_percent=idx / N, stage = YmirStage.TASK)
+            write_ymir_monitor_process(cfg, task='infer', naive_stage_percent=idx / N, stage=YmirStage.TASK)
+
+    if WORLD_SIZE > 1:
+        infer_result = collect_results_gpu(infer_result, len(images))
 
     rw.write_infer_result(infer_result=infer_result)
     write_ymir_monitor_process(cfg, task='infer', naive_stage_percent=1.0, stage=YmirStage.POSTPROCESS)

det-mmdetection-tmi/ymir_mining_aldd.py

@@ -2,12 +2,13 @@
 
 import torch
 from easydict import EasyDict as edict
-from mining_base import ALDDMining
 from mmcv.parallel import collate, scatter
+from ymir_exc.util import get_merged_config
+
+from mining_base import ALDDMining
 from mmdet.datasets import replace_ImageToTensor
 from mmdet.datasets.pipelines import Compose
 from mmdet.models.detectors import YOLOX
-from ymir_exc.util import get_merged_config
 from ymir_infer import YmirModel
 from ymir_mining_random import RandomMiner
 

det-mmdetection-tmi/ymir_train.py

@@ -5,9 +5,11 @@
 import sys
 
 from easydict import EasyDict as edict
-from mmdet.utils.util_ymir import get_best_weight_file, write_ymir_training_result
-from ymir_exc import monitor
-from ymir_exc.util import YmirStage, find_free_port, get_merged_config, write_ymir_monitor_process
+from ymir_exc.util import (YmirStage, find_free_port, get_merged_config,
+                           write_ymir_monitor_process)
+
+from mmdet.utils.util_ymir import (get_best_weight_file,
+                                   write_ymir_training_result)
 
 
 def main(cfg: edict) -> int:

docs/cn/docker_images/det-nanodet-tmi.md

@@ -10,6 +10,7 @@
 ## 镜像地址
 ```
 youdaoyzbx/ymir-executor:ymir2.0.0-nanodet-cu111-tmi
+youdaoyzbx/ymir-executor:ymir2.0.2-nanodet-cu111-tmi
 ```
 
 ## 性能说明

docs/cn/docker_images/det-yolov5-tmi.md

@@ -1,5 +1,11 @@
 # yolov5 镜像说明文档
 
+
+## 仓库地址
+
+> 参考[ultralytics/yolov5](https://github.com/ultralytics/yolov5)
+- [modelai/ymir-executor-fork](https://github.com/modelai/ymir-executor-fork/tree/master/det-yolov5-tmi)
+
 ## 镜像地址
 
 ```