Improving experiment configuration

pcdet/config.py

@@ -1,7 +1,9 @@
 from pathlib import Path
+from typing import List
 
 import yaml
 from easydict import EasyDict
+from runcon import Config
 
 
 def log_config_to_file(cfg, pre='cfg', logger=None):
@@ -80,6 +82,74 @@ def cfg_from_yaml_file(cfg_file, config):
     return config
 
 
+def dict_representer(dumper, ed: EasyDict):
+    return dumper.represent_mapping('tag:yaml.org,2002:map', ed)
+
+
+yaml.add_representer(EasyDict, dict_representer)
+
+
+def cfg_to_yaml_file(config, cfg_file):
+    with open(cfg_file, 'w') as f:
+        yaml.dump(config, f, default_flow_style=False)
+
+
+def check_recursive_dict_type(cfg, type):
+    if isinstance(cfg, dict):
+        if not isinstance(cfg, type):
+            raise ValueError(
+                "The config is of type %s but was expected to be of type %s!\n%s"
+                % (type(cfg), type, cfg)
+            )
+        for k in cfg:
+            check_recursive_dict_type(cfg[k], type)
+
+
+def ed_to_rc(ed_cfg: EasyDict) -> Config:
+    check_recursive_dict_type(ed_cfg, EasyDict)
+    rc_cfg = Config(ed_cfg)
+    check_recursive_dict_type(rc_cfg, Config)
+    return rc_cfg
+
+
+def rc_to_ed(rc_cfg: Config) -> EasyDict:
+    check_recursive_dict_type(rc_cfg, Config)
+    ed_cfg = EasyDict()
+    for k in rc_cfg:
+        if isinstance(rc_cfg[k], Config):
+            ed_cfg[k] = rc_to_ed(rc_cfg[k])
+        else:
+            ed_cfg[k] = rc_cfg[k]
+    check_recursive_dict_type(ed_cfg, EasyDict)
+    return ed_cfg
+
+
+def modify_rc_cfg(cfg: Config, modify_cfgs: List[Path]) -> Config:
+    from copy import deepcopy
+    cfg = deepcopy(cfg)
+    for m in modify_cfgs:
+        m = cfg_from_yaml_file(m, EasyDict())
+        cfg.rupdate(ed_to_rc(m))
+    cfg = cfg.resolve_transforms()
+    return cfg
+
+
+def create_cfg_from_sets(
+        cfg_file: Path,
+        modify_cfgs: List[Path],
+        set_cfgs: List[str],
+        cfg: EasyDict = None,
+) -> EasyDict:
+    if cfg is None:
+        cfg = EasyDict()
+    cfg_from_yaml_file(cfg_file, cfg)
+    cfg = ed_to_rc(cfg)
+    cfg = modify_rc_cfg(cfg, modify_cfgs)
+    cfg = rc_to_ed(cfg)
+    cfg_from_list(set_cfgs, cfg)
+    return cfg
+
+
 cfg = EasyDict()
 cfg.ROOT_DIR = (Path(__file__).resolve().parent / '../').resolve()
 cfg.LOCAL_RANK = 0

requirements.txt

@@ -10,4 +10,5 @@ tqdm
 torchvision
 SharedArray
 opencv-python
-pyquaternion
+pyquaternion
+runcon

tools/cfgs/dataset_configs/modifiers/waymo/multiframe.yaml

@@ -0,0 +1,64 @@
+SEQUENCE_CONFIG:
+    ENABLED: True
+    SAMPLE_OFFSET: [-3, 0]
+
+TRAIN_WITH_SPEED: True 
+
+
+DATA_AUGMENTOR:
+    AUG_CONFIG_LIST:
+        - NAME: gt_sampling
+          USE_ROAD_PLANE: False
+          DB_INFO_PATH:
+              - waymo_processed_data_v0_5_0_waymo_dbinfos_train_sampled_1_multiframe_-4_to_0.pkl
+
+          USE_SHARED_MEMORY: False  # set it to True to speed up (it costs about 50GB? shared memory)
+          DB_DATA_PATH:
+              - waymo_processed_data_v0_5_0_gt_database_train_sampled_1_multiframe_-4_to_0_global.npy
+
+          PREPARE: {
+             filter_by_min_points: ['Vehicle:5', 'Pedestrian:5', 'Cyclist:5'],
+             filter_by_difficulty: [-1],
+          }
+
+          SAMPLE_GROUPS: ['Vehicle:15', 'Pedestrian:10', 'Cyclist:10']
+          NUM_POINT_FEATURES: 6
+          REMOVE_EXTRA_WIDTH: [0.0, 0.0, 0.0]
+          LIMIT_WHOLE_SCENE: True
+
+          FILTER_OBJ_POINTS_BY_TIMESTAMP: True 
+          TIME_RANGE: [0.3, 0.0]  # 0.3s-0.0s indicates 4 frames 
+
+        - NAME: random_world_flip
+          ALONG_AXIS_LIST: ['x', 'y']
+
+        - NAME: random_world_rotation
+          WORLD_ROT_ANGLE: [-0.78539816, 0.78539816]
+
+        - NAME: random_world_scaling
+          WORLD_SCALE_RANGE: [0.95, 1.05]
+
+
+POINT_FEATURE_ENCODING:
+    used_feature_list: ['x', 'y', 'z', 'intensity', 'elongation', 'timestamp']
+    src_feature_list: ['x', 'y', 'z', 'intensity', 'elongation', 'timestamp']
+
+
+DATA_PROCESSOR:
+    - NAME: mask_points_and_boxes_outside_range
+      REMOVE_OUTSIDE_BOXES: True
+      USE_CENTER_TO_FILTER: True
+
+    - NAME: shuffle_points
+      SHUFFLE_ENABLED: {
+        'train': True,
+        'test': True
+      }
+
+    - NAME: transform_points_to_voxels
+      VOXEL_SIZE: [0.1, 0.1, 0.15]
+      MAX_POINTS_PER_VOXEL: 5
+      MAX_NUMBER_OF_VOXELS: {
+        'train': 180000,
+        'test': 400000
+      }

tools/cfgs/waymo_models/modifiers/mppnet/16frames.yaml

@@ -0,0 +1,16 @@
+DATA_CONFIG:
+    SEQUENCE_CONFIG:
+        SAMPLE_OFFSET: [-15,0]
+
+MODEL:
+    ROI_HEAD:
+        TRANS_INPUT: 64
+
+        ROI_GRID_POOL:
+            MLPS: [[64,64]]
+            POOL_RADIUS: [0.8]
+            NSAMPLE: [16]
+
+        Transformer:
+            num_frames: 16
+            sequence_stride: 4

tools/cfgs/waymo_models/modifiers/mppnet/4frames.yaml

@@ -0,0 +1,15 @@
+DATA_CONFIG:
+    SEQUENCE_CONFIG:
+        SAMPLE_OFFSET: [-3,0]
+
+MODEL:
+    ROI_HEAD:
+        TRANS_INPUT: 256
+
+        ROI_GRID_POOL:
+            MLPS: [[128,128], [128,128]]
+            POOL_RADIUS: [0.8, 1.6]
+            NSAMPLE: [16, 16]
+
+        Transformer:
+            num_frames: 4

tools/cfgs/waymo_models/modifiers/mppnet/e2e_memorybank_inference.yaml

@@ -0,0 +1,93 @@
+DATA_CONFIG:
+    _BASE_CONFIG_: cfgs/dataset_configs/modifiers/waymo/multiframe.yaml
+    _TRANSFORM:
+    - name: remove_element
+      target: ROI_BOXES_PATH
+    - name: remove_element
+      target: USE_PREDBOX
+
+    SAMPLED_INTERVAL: {
+        'train': 5,
+        'test': 1
+    }
+
+    POINT_FEATURE_ENCODING:
+        encoding_type: absolute_coordinates_encoding
+        used_feature_list: ['x', 'y', 'z', 'intensity', 'elongation', 'time']
+        src_feature_list: ['x', 'y', 'z', 'intensity', 'elongation', 'time']
+
+MODEL:
+    NAME: MPPNetE2E
+
+    VFE:
+        NAME: DynMeanVFE
+
+    BACKBONE_3D:
+        NAME: VoxelResBackBone8x
+
+    MAP_TO_BEV:
+        NAME: HeightCompression
+        NUM_BEV_FEATURES: 256
+
+    BACKBONE_2D:
+        NAME: BaseBEVBackbone 
+        NUM_FRAME: 2
+        LAYER_NUMS: [5, 5]
+        LAYER_STRIDES: [1, 2]
+        NUM_FILTERS: [128, 256]
+        UPSAMPLE_STRIDES: [1, 2]
+        NUM_UPSAMPLE_FILTERS: [256, 256]
+
+    DENSE_HEAD:
+        NAME: CenterHead
+        CLASS_AGNOSTIC: False
+
+        CLASS_NAMES_EACH_HEAD: [
+            ['Vehicle', 'Pedestrian', 'Cyclist']
+        ]
+
+        SHARED_CONV_CHANNEL: 64
+        USE_BIAS_BEFORE_NORM: True
+        NUM_HM_CONV: 2
+        SEPARATE_HEAD_CFG:
+            HEAD_ORDER: ['center', 'center_z', 'dim', 'rot','vel']
+            HEAD_DICT: {
+                'center': {'out_channels': 2, 'num_conv': 2},
+                'center_z': {'out_channels': 1, 'num_conv': 2},
+                'dim': {'out_channels': 3, 'num_conv': 2},
+                'rot': {'out_channels': 2, 'num_conv': 2},
+                'vel': {'out_channels': 2, 'num_conv': 2},
+
+            }
+
+        TARGET_ASSIGNER_CONFIG:
+            FEATURE_MAP_STRIDE: 8
+            NUM_MAX_OBJS: 500
+            GAUSSIAN_OVERLAP: 0.1
+            MIN_RADIUS: 2
+
+        LOSS_CONFIG:
+            LOSS_WEIGHTS: {
+                'cls_weight': 1.0,
+                'loc_weight': 2.0,
+                'code_weights': [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.2, 0.2]
+            }
+
+        POST_PROCESSING:
+            SCORE_THRESH: 0.1
+            POST_CENTER_LIMIT_RANGE: [-75.2, -75.2, -2, 75.2, 75.2, 4]
+            MAX_OBJ_PER_SAMPLE: 500
+            NMS_CONFIG:
+                NMS_TYPE: nms_gpu
+                NMS_THRESH: 0.7
+                NMS_PRE_MAXSIZE: 4096
+                NMS_POST_MAXSIZE: 500
+
+    ROI_HEAD:
+        NAME: MPPNetHeadE2E
+        USE_BOX_ENCODING:
+            NORM_T0: True
+            ALL_YAW_T0: True
+
+        Transformer:
+            sequence_stride: 1 #16frame using 4

tools/cfgs/waymo_models/mppnet_4frames.yaml → tools/cfgs/waymo_models/mppnet.yaml

@@ -45,9 +45,9 @@ DATA_CONFIG:
             }
 
     POINT_FEATURE_ENCODING: {
-        encoding_type: absolute_coordinates_encoding, 
-        used_feature_list: ['x', 'y', 'z', 'intensity', 'elongation','time'],
-        src_feature_list: ['x', 'y', 'z', 'intensity', 'elongation','time'],       
+        encoding_type: absolute_coordinates_encoding,
+        used_feature_list: ['x', 'y', 'z', 'intensity', 'elongation', 'time'],
+        src_feature_list: ['x', 'y', 'z', 'intensity', 'elongation', 'time'],
         }
 
 
@@ -75,13 +75,16 @@ MODEL:
             POOL_METHOD: max_pool
 
 
+        _TRANSFORM:
+            name: copy
+            src: TRANS_INPUT
+            dest: Transformer.hidden_dim
         Transformer:
             num_lidar_points: 128
             num_proxy_points: 64  # GRID_SIZE*GRID_SIZE*GRID_SIZE
             pos_hidden_dim: 64
             enc_layers: 3
             dim_feedforward: 512
-            hidden_dim: 256 #equal to ROI_HEAD.TRANS_INPUT
             dropout: 0.1
             nheads: 4
             pre_norm: False