update ddp

.gitignore

@@ -0,0 +1,11 @@
+.git
+_debug/
+__pycache__/
+pretrained/
+checkpoint/
+*.pth
+dataset/debug_tools.py
+debugger/
+dataset/
+output_all/
+tf_logs/

README.md

@@ -1,6 +1,13 @@
 # monocon_rp (2022.08.10)
 此项目用于复现monocon 
 源码地址：https://github.com/chenzihao008/monocon-pytorch.git
+## 修改部分
+### 将训练源码修改为DDP训练,使用mp.spwan启动，单机单卡、单机多卡均可使用
+- [修改了以下文件：]
+- engine\base_engine.py
+- engine\monocon_engine.py
+- train.py
+### 存在dataloader多进程读取数据占用显存问题
 ## environment
 - [显卡] 3080*1 10G
 - [CUDA] 10.1

changelog.MD

@@ -0,0 +1,17 @@
+## Changelog
+
+### v1.0.3 (2022.09.10)
+- Made ```torch.backend.cudnn.benchmark``` option configurable in ```monocon_config.py```. (**Issue #2**)
+  ```python
+  _C.USE_BENCHMARK = True
+  ```
+- Some code has been modified and added to enable video inference using kitti raw dataset.
+
+### v1.0.2 (2022.08.30)
+- A Transform class ```Resize3D``` for resize augmentation has been added. (**Issue #1**)
+
+### v1.0.1 (2022.08.29)
+- Visualization functions for 2D Boxes, 3D Boxes, and BEV have been added.
+
+### v1.0.0 (2022.08.28)
+- This repository is released.

config.yaml

@@ -0,0 +1,41 @@
+DATA:
+  BATCH_SIZE: 2
+  FILTER:
+    MAX_DEPTH: 65
+    MAX_OCCLUSION: 2
+    MAX_TRUNCATION: 0.5
+    MIN_DEPTH: 2
+    MIN_HEIGHT: 25
+  NUM_WORKERS: 0
+  ROOT: dataset
+  TEST_SPLIT: val
+  TRAIN_SPLIT: train
+DESCRIPTION: MonoCon Default Configuration
+GPU_ID: 0
+MODEL:
+  BACKBONE:
+    IMAGENET_PRETRAINED: false
+    NUM_LAYERS: 34
+  HEAD:
+    MAX_OBJS: 30
+    NUM_CLASSES: 3
+OUTPUT_DIR: ''
+PERIOD:
+  EVAL_PERIOD: 1
+  LOG_PERIOD: 50
+SEED: 1891176477
+SOLVER:
+  CLIP_GRAD:
+    ENABLE: true
+    MAX_NORM: 35
+    NORM_TYPE: 2.0
+  OPTIM:
+    LR: 0.000225
+    NUM_EPOCHS: 5
+    WEIGHT_DECAY: 1.0e-05
+  SCHEDULER:
+    ENABLE: true
+TRAIN_TYPE: DDP
+USE_BENCHMARK: false
+VERSION: v1.0.3
+

config/monocon_configs.py

@@ -0,0 +1,65 @@
+from yacs.config import CfgNode as CN
+
+
+_C = CN()
+
+_C.VERSION = 'v1.0.3'
+_C.DESCRIPTION = "MonoCon Default Configuration"
+
+_C.OUTPUT_DIR = ""                               # Output Directory
+_C.SEED = -1                                     # -1: Random Seed Selection
+_C.GPU_ID = 0                                    # Index of GPU to use
+
+_C.USE_BENCHMARK = False                          # Value of 'torch.backends.cudnn.benchmark' and 'torch.backends.cudnn.enabled'
+
+_C.TRAIN_TYPE = 'DDP'
+# Data
+_C.DATA = CN()
+_C.DATA.ROOT = 'dataset'                  # KITTI Root
+_C.DATA.BATCH_SIZE = 8
+_C.DATA.NUM_WORKERS = 0
+_C.DATA.TRAIN_SPLIT = 'train'
+_C.DATA.TEST_SPLIT = 'val'
+
+_C.DATA.FILTER = CN()
+_C.DATA.FILTER.MIN_HEIGHT = 25
+_C.DATA.FILTER.MIN_DEPTH = 2
+_C.DATA.FILTER.MAX_DEPTH = 65
+_C.DATA.FILTER.MAX_TRUNCATION = 0.5
+_C.DATA.FILTER.MAX_OCCLUSION = 2
+
+
+# Model
+_C.MODEL = CN()
+
+_C.MODEL.BACKBONE = CN()
+_C.MODEL.BACKBONE.NUM_LAYERS = 34
+_C.MODEL.BACKBONE.IMAGENET_PRETRAINED = False   #不用预训练的backbone效果肯定会好一点吧
+
+_C.MODEL.HEAD = CN()
+_C.MODEL.HEAD.NUM_CLASSES = 3
+_C.MODEL.HEAD.MAX_OBJS = 30
+
+
+# Optimization
+_C.SOLVER = CN()
+
+_C.SOLVER.OPTIM = CN()
+_C.SOLVER.OPTIM.LR = 2.25E-04
+_C.SOLVER.OPTIM.WEIGHT_DECAY = 1E-05
+_C.SOLVER.OPTIM.NUM_EPOCHS = 5        # Max Training Epochs
+
+_C.SOLVER.SCHEDULER = CN()
+_C.SOLVER.SCHEDULER.ENABLE = True
+
+_C.SOLVER.CLIP_GRAD = CN()
+_C.SOLVER.CLIP_GRAD.ENABLE = True
+_C.SOLVER.CLIP_GRAD.NORM_TYPE = 2.0
+_C.SOLVER.CLIP_GRAD.MAX_NORM = 35 
+
+
+# Period
+_C.PERIOD = CN()
+_C.PERIOD.EVAL_PERIOD = 1                      # In Epochs / Set -1 if you don't want validation
+_C.PERIOD.LOG_PERIOD = 50                       # In Steps
+