search

megvii-research / MCTrack

This is the offical implementation of the paper "MCTrack: A Unified 3D Multi-Object Tracking Framework for Autonomous Driving"
51 stars 1 forks source link

为什么我在nusences上的测试结果达不到0.763 #9

Open ZZC-CN opened 3 hours ago

ZZC-CN commented 3 hours ago

image 我的配置文件为

------------------------------------------------------------------------

Copyright (c) 2024 megvii-research. All Rights Reserved.

------------------------------------------------------------------------

DATASET: "nuscenes"

dataset info

SPLIT: "val" # val, test DETECTOR: centerpoint # centerpoint, largekernel

DATASET_ROOT: "s3://wangxiyang/open_datasets/nuscenes/raw_data/"

DATASET_ROOT: "/workspace/mmdetection3d-main_old/data/nuscenes" # 开源用这个 DETECTIONS_ROOT: "/workspace/mmdetection3d-main_old/data/nuscenes/" SAVE_PATH: "results/" TRACKING_MODE: 'ONLINE' # ONLINE / GLOBAL IS_RV_MATCHING: False # True / False FRAME_RATE: 2 CATEGORY_LIST: ["car", "pedestrian", "bicycle", "motorcycle", "bus", "trailer", "truck"] CATEGORY_MAP_TO_NUMBER : { "car": 0, "pedestrian": 1, "bicycle": 2, "motorcycle": 3, "bus": 4, "trailer": 5, "truck": 6, }

MATCHING: BEV:

Predict, BackPredict, Fusion

COST_STATE: {0: "BackPredict", 1: "BackPredict", 2: "Predict", 3: "Fusion", 4: "Predict", 5: "BackPredict", 6: "BackPredict"} 
# iou_3d, giou_3d, dist_3d, ro_gdiou_3d
COST_MODE: {0: "RO_GDIOU_3D", 1: "RO_GDIOU_3D", 2: "RO_GDIOU_3D", 3: "RO_GDIOU_3D", 4: "RO_GDIOU_3D", 5: "RO_GDIOU_3D", 6: "RO_GDIOU_3D"} 
 # Hungarian, Greedy
MATCHING_MODE: Hungarian

RV:

Predict, BackPredict, Fusion

COST_STATE: {0: "Predict", 1: "Predict", 2: "Predict", 3: "Predict", 4: "Predict", 5: "Predict", 6: "Predict"} 
# IOU_2D, GIOU_2D, DIOU_2D, SDIOU_2D
COST_MODE: {0: "SDIOU_2D", 1: "SDIOU_2D", 2: "SDIOU_2D", 3: "SDIOU_2D", 4: "SDIOU_2D", 5: "SDIOU_2D", 6: "SDIOU_2D"} 
 # Hungarian, Greedy
MATCHING_MODE : Greedy  # Hungarian, Greedy

----------------threshold------------

THRESHOLD: INPUT_SCORE: ONLINE: {0: 0.15, 1: 0.16, 2: 0.20, 3: 0.15, 4: 0.16, 5: 0.17, 6: 0.0} OFFLINE: {0: 0.15, 1: 0.16, 2: 0.20, 3: 0.15, 4: 0.16, 5: 0.17, 6: 0.0} COST_STATE_PREDICT_RATION: {0: 1.0, 1: 1.0, 2: 1.0, 3: 0.5, 4: 1.0, 5: 1.0, 6: 1.0} NMS_THRE: {0: 0.04, 1: 0.08, 2: 0.08, 3: 0.06, 4: 0.08, 5: 0.12, 6: 0.05}
BEV: COST_THRE: {0: 1.10, 1: 2.06, 2: 2.00, 3: 2.06, 4: 1.60, 5: 1.26, 6: 1.16} WEIGHT_RO_GDIOU: { 0: {"w1": 0.5, "w2": 1.5}, 1: {"w1": 1.0, "w2": 1.0}, 2: {"w1": 1.0, "w2": 1.0}, 3: {"w1": 1.0, "w2": 1.0}, 4: {"w1": 1.0, "w2": 1.0}, 5: {"w1": 1.0, "w2": 1.0}, 6: {"w1": 1.0, "w2": 1.0}, } RV: COST_THRE: {0: -0.3, 1: -0.3, 2: -0.3, 3: -0.3, 4: -0.3, 5: -0.3, 6: -0.3}

TRAJECTORY_THRE: CACHE_BBOX_LENGTH: {0: 30, 1: 30, 2: 30, 3: 30, 4: 30, 5: 30, 6: 30} PREDICT_BBOX_LENGTH: {0: 17, 1: 7, 2: 13, 3: 22, 4: 14, 5: 7, 6: 22} MAX_UNMATCH_LENGTH: {0: 0, 1: 1, 2: 1, 3: 1, 4: 1, 5: 1, 6: 1} CONFIRMED_TRACK_LENGTH: {0: 1, 1: 1, 2: 1, 3: 1, 4: 1, 5: 1, 6: 1} DELET_OUT_VIEW_LENGTH: {0: 0, 1: 0, 2: 0, 3: 0, 4: 0, 5: 0, 6: 0}
CONFIRMED_DET_SCORE: {0: 0.7, 1: 0.7, 2: 0.7, 3: 0.7, 4: 0.7, 5: 0.7, 6: 0.7} IS_FILTER_PREDICT_BOX: {0: -1, 1: -1, 2: -1, 3: -1, 4: -1, 5: -1, 6: -1} CONFIRMED_MATCHED_SCORE: {0: 0.3, 1: 0.3, 2: 0.3, 3: 0.3, 4: 0.3, 5: 0.3, 6: 0.3}

KALMAN_FILTER_POSE: MOTION_MODE: {0: "CV", 1: "CV", 2: "CV", 3: "CV", 4: "CV", 5: "CV", 6: "CV"} # CV, CA, CTRA CV: N: 4 # State Dimension M: 4 # Measure Dimension NOISE: 0: P: [1.0, 1.0, 10.0, 10.0] Q: [0.5, 0.5, 1.5, 1.5] R: [0.7, 0.7, 0.5, 0.5] 1: P: [1.0, 1.0, 10.0, 10.0] Q: [1.5, 1.5, 1.5, 1.5] R: [2.0, 2.0, 3.5, 3.5] 2: P: [1.0, 1.0, 1.0, 1.0] Q: [0.3, 0.3, 1.0, 1.0] R: [0.1, 0.1, 1.0, 1.0] 3: P: [1.0, 1.0, 10.0, 10.0] Q: [0.5, 0.5, 4.0, 4.0] R: [0.1, 0.1, 0.1, 0.1] 4: P: [100.0, 100.0, 100.0, 100.0] Q: [0.5, 0.5, 1.5, 1.5] R: [1.5, 1.5, 500, 500] 5: P: [1.0, 1.0, 10.0, 10.0] Q: [0.3, 0.3, 0.1, 0.1] R: [2.0, 2.0, 2.5, 2.5] 6: P: [1.0, 1.0, 10.0, 10.0] Q: [0.1, 0.1, 2.0, 2.0] R: [1.5, 1.5, 4.0, 4.0] CA: N: 6 # State Dimension M: 2 # Measure Dimension NOISE: 0: P: [1.0, 1.0, 1.0, 1.0, 1.0, 1.0] Q: [2.0, 2.0, 1.0, 0.5, 1.0, 1.5] R: [0.5, 0.5] 1: P: [1.0, 1.0, 1.0, 1.0, 1.0, 1.0] Q: [2.0, 2.0, 1.0, 0.5, 1.0, 1.5] R: [0.5, 0.5] 2: P: [1.0, 1.0, 1.0, 1.0, 1.0, 1.0] Q: [2.0, 2.0, 1.0, 0.5, 1.0, 1.5] R: [0.5, 0.5] 3: P: [1.0, 1.0, 1.0, 1.0, 1.0, 1.0] Q: [2.0, 2.0, 1.0, 0.5, 1.0, 1.5] R: [0.5, 0.5] 4: P: [1.0, 1.0, 1.0, 1.0, 1.0, 1.0] Q: [2.0, 2.0, 1.0, 0.5, 1.0, 1.5] R: [0.5, 0.5] 5: P: [1.0, 1.0, 1.0, 1.0, 1.0, 1.0] Q: [2.0, 2.0, 1.0, 0.5, 1.0, 1.5] R: [0.5, 0.5] 6: P: [1.0, 1.0, 1.0, 1.0, 1.0, 1.0] Q: [2.0, 2.0, 1.0, 0.5, 1.0, 1.5] R: [0.5, 0.5] CTRA: N: 6 # State Dimension M: 2 # Measure Dimension NOISE: 0: P: [1.0, 1.0, 1.0, 1.0, 1.0, 1.0] Q: [2.0, 2.0, 1.0, 0.5, 1.0, 1.5] R: [0.5, 0.5] 1: P: [1.0, 1.0, 1.0, 1.0, 1.0, 1.0] Q: [2.0, 2.0, 1.0, 0.5, 1.0, 1.5] R: [0.5, 0.5] 2: P: [1.0, 1.0, 1.0, 1.0, 1.0, 1.0] Q: [2.0, 2.0, 1.0, 0.5, 1.0, 1.5] R: [0.5, 0.5] 3: P: [1.0, 1.0, 1.0, 1.0, 1.0, 1.0] Q: [2.0, 2.0, 1.0, 0.5, 1.0, 1.5] R: [0.5, 0.5] 4: P: [1.0, 1.0, 1.0, 1.0, 1.0, 1.0] Q: [2.0, 2.0, 1.0, 0.5, 1.0, 1.5] R: [0.5, 0.5] 5: P: [1.0, 1.0, 1.0, 1.0, 1.0, 1.0] Q: [2.0, 2.0, 1.0, 0.5, 1.0, 1.5] R: [0.5, 0.5] 6: P: [1.0, 1.0, 1.0, 1.0, 1.0, 1.0] Q: [2.0, 2.0, 1.0, 0.5, 1.0, 1.5] R: [0.5, 0.5]

KALMAN_FILTER_SIZE: MOTION_MODE: {0: "CV", 1: "CV", 2: "CV", 3: "CV", 4: "CV", 5: "CV", 6: "CV"} CV: N: 4 # State Dimension M: 2 # Measure Dimension NOISE: 0: P: [1.0, 1.0, 10.0, 10.0] Q: [0.5, 0.5, 1.5, 1.5] R: [2.0, 2.0] 1: P: [1.0, 1.0, 10.0, 10.0] Q: [0.5, 0.5, 1.5, 1.5] R: [2.0, 2.0] 2: P: [1.0, 1.0, 10.0, 10.0] Q: [0.5, 0.5, 1.5, 1.5] R: [2.0, 2.0] 3: P: [1.0, 1.0, 10.0, 10.0] Q: [0.5, 0.5, 1.5, 1.5] R: [2.0, 2.0] 4: P: [1.0, 1.0, 10.0, 10.0] Q: [0.5, 0.5, 1.5, 1.5] R: [2.0, 2.0] 5: P: [1.0, 1.0, 10.0, 10.0] Q: [0.5, 0.5, 1.5, 1.5] R: [2.0, 2.0] 6: P: [1.0, 1.0, 10.0, 10.0] Q: [0.5, 0.5, 1.5, 1.5] R: [2.0, 2.0]

KALMAN_FILTER_YAW: MOTION_MODE: {0: "CV", 1: "CV", 2: "CV", 3: "CV", 4: "CV", 5: "CV", 6: "CV"} CV: N: 2 # State Dimension M: 2 # Measure Dimension NOISE: 0: P: [0.1, 0.1] Q: [0.1, 0.1] R: [0.2, 5.0] 1: P: [0.1, 0.1] Q: [0.1, 0.1] R: [0.2, 5.0] 2: P: [0.1, 0.1] Q: [0.1, 0.1] R: [0.2, 5.0] 3: P: [0.1, 0.1] Q: [0.1, 0.1] R: [0.2, 5.0] 4: P: [0.1, 0.1] Q: [0.1, 0.1] R: [0.2, 5.0] 5: P: [0.1, 0.1] Q: [0.1, 0.1] R: [0.2, 5.0] 6: P: [0.1, 0.1] Q: [0.1, 0.1] R: [0.2, 5.0]

KALMAN_FILTER_RVBOX: MOTION_MODE: {0: "CV"} CV: N: 8 # State Dimension M: 4 # Measure Dimension NOISE: 0: P: [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0] Q: [0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5, 0.5] R: [2.0, 2.0, 2.0, 2.0]

wodex1nhaoIeng commented 2 hours ago

hi，我们提交测试集的时候并没有用centerpoint，你需要把检测器换成largekernel