def exp(Experiment_name, hyp_param, df):
PHASE = "training"
Basline_Name = "AB3DMOT"
DATASET_NAME = "kitti"
debug_level = paraser_args.debug_level
SCRIPT_DIR = "/usr/stud/kaddah/ciwt_gd"
DATASET_DIR = "/usr/stud/kaddah/storage/datasets/"
DATASET = DATASET_DIR + DATASET_NAME + "/{}".format(PHASE)
PREPROC = DATASET_DIR + DATASET_NAME + "/{}/preproc".format(PHASE)
OUT_DIR = DATASET_DIR + DATASET_NAME + "/{}/{}".format(
PHASE, paraser_args.tracking_output)
# Possible modes: 'detection', 'detection_shape'
# The latter is considered obsolete; this option is still here for eval purposes
MODE = "detection" # detection_shape
CIWT = "/usr/stud/kaddah/ciwt_gd/build/apps/CIWTApp"
# Specify detectors (should correspond to detector dir names and config names)
# DETS=['det_02_regionlets', 'det_02_3DOP', 'pointrcnn']
DETS = ['pointrcnn']
if DATASET_NAME == "kitti":
if PHASE == "training" or PHASE == "validation":
end_frames = [
"153", "446", "232", "143", "313", "296", "269", "799", "389",
"802", "293", "372", "77", "339", "105", "375", "208", "144",
"338", "1058", "836"
]
SEQUENCES = [2, 6, 7, 8, 10, 13, 14, 16, 18]
else:
end_frames = [
"464", "146", "242", "256", "420", "808", "113", "214", "164",
"348", "1175", "773", "693", "151", "849", "700", "509", "304",
"179", "403", "172", "202", "435", "429", "315", "175", "169",
"84", "174"
]
SEQUENCES = range(29)
# Path to your binary
# e.sequence_name
# ['0002', '0006', '0007', '0008', '0010', '0013', '0014', '0016', '0018']
# e.n_frames
# [233, 270, 800, 390, 294, 340, 106, 209, 339]
# Exec tracker for each detection set
for DET in DETS:
# Exec tracker for each sequence
for SEQ in SEQUENCES:
SEQNAME = "%04d" % (SEQ)
print("Process: {}".format(SEQNAME))
IM_LEFT = "{}/image_02/{}/%06d.png".format(DATASET, SEQNAME)
IM_RIGHT = "{}/image_03/{}/%06d.png".format(DATASET, SEQNAME)
DISP = "{}/disparity/disp_{}/%06d_left.png".format(
PREPROC, SEQNAME)
END_FRAME = end_frames[SEQ]
CALIB = "{}/calib/{}.txt".format(DATASET, SEQNAME)
PROP = "{}/cached_proposals/{}/%06d.bin".format(PREPROC, SEQNAME)
OUT_DIR_DET = "{}/{}".format(OUT_DIR, DET)
DETPATH = "{}/detection/{}/{}/%06d.txt".format(
PREPROC, DET, SEQNAME)
DETCFG = "{}/cfg/{}.cfg".format(SCRIPT_DIR, DET)
if os.path.exists(SCRIPT_DIR + "/results/" + Experiment_name):
shutil.rmtree(SCRIPT_DIR + "/results/" + Experiment_name)
shutil.copytree(SCRIPT_DIR + "/results/ready_results_baseline",
SCRIPT_DIR + "/results/" + Experiment_name)
args = [
"--config",
str(DETCFG), "--left_image_path",
str(IM_LEFT), "--end_frame",
str(END_FRAME), "--right_image_path",
str(IM_RIGHT), "--left_disparity_path",
str(DISP), "--calib_path",
str(CALIB), "--object_proposal_path",
str(PROP), "--detections_path",
str(DETPATH), "--output_dir",
str(OUT_DIR_DET), "--tracking_mode",
str(MODE), "--sequence_name",
str(SEQNAME), "--debug_level",
str(debug_level), "--is_oriented",
str(1), "--enable_coupling",
str(1), "--check_exit_to2D_projeciton",
str(0), "--association_score_is_3DIoU",
str(1), "--association_appearance_model_weight",
str(hyp_param["association_appearance_model_weight"]),
"--IoU3D_model_weight",
str(hyp_param["IoU3D_model_weight"]),
"--association_weight_distance_from_camera_param",
str(hyp_param["association_weight_distance_from_camera_param"]
), "--gaiting_appearance_threshold",
str(hyp_param["gaiting_appearance_threshold"]),
"--gaiting_IOU_threshold",
str(hyp_param["gaiting_IOU_threshold"]),
"--gaiting_mh_distance_threshold",
str(hyp_param["gaiting_mh_distance_threshold"]),
"--gaiting_min_association_score",
str(hyp_param["gaiting_min_association_score"]),
"--gaiting_size_2D",
str(hyp_param["gaiting_size_2D"]), "--tracking_e1",
str(hyp_param["tracking_e1"]), "--tracking_e3",
str(hyp_param["tracking_e3"]), "--tracking_e4",
str(hyp_param["tracking_e4"]),
"--id_handling_overlap_threshold",
str(hyp_param["id_handling_overlap_threshold"]),
"--tracking_exp_decay",
str(hyp_param["tracking_exp_decay"]), "--max_hole_size",
str(hyp_param["max_hole_size"]),
"--tracking_temporal_window_size",
str(hyp_param["tracking_temporal_window_size"]),
"--hole_penalty_decay_parameter",
str(hyp_param["hole_penalty_decay_parameter"]),
"--min_observations_needed_to_init_hypothesis",
str(hyp_param["min_observations_needed_to_init_hypothesis"]),
"--tracking_single_detection_hypo_threshold",
str(hyp_param["tracking_single_detection_hypo_threshold"])
# ,
# "--kf_2d_observation_noise", str(hpy["kf_2d_observation_noise"]) ,
# "--kf_2d_initial_variance", str(hpy["kf_2d_initial_variance"]) ,
# "--kf_2d_system_noise", str(hpy["kf_2d_system_noise"]) ,
# "--kf_3d_observation_noise", str(hpy["kf_3d_observation_noise"]) ,
# "--kf_3d_initial_variance",str(hpy["kf_3d_initial_variance"]) ,
# "--kf_3d_system_noise",str(hpy["kf_3d_system_noise"]) ,
# "--kf_system_pos_variance_car_x",str(hpy["kf_system_pos_variance_car_x"]) ,
# "--kf_system_pos_variance_car_z",str(hpy["kf_system_pos_variance_car_z"]) ,
# "--kf_system_velocity_variance_car_x",str(hpy["kf_system_velocity_variance_car_x"]) ,
# "--kf_system_velocity_variance_car_z",str(hpy["kf_system_velocity_variance_car_z"]) ,
# "--kf_orientation_obs_noise",str(hpy["kf_orientation_obs_noise"])
]
subprocess.run([CIWT] + args)
for DET in DETS:
shutil.move(
OUT_DIR + "/" + DET + "/data", SCRIPT_DIR + "/results" + "/" +
Experiment_name + "/CIWT_Pointrcnn/val/")
PHASE = "training"
# Determine sequences here
if PHASE == "training":
# SEQUENCES = [0,1,3,4,5,9,11,12,15,17,19,20]
# elif PHASE == "validation":
SEQUENCES = [2, 6, 7, 8, 10, 13, 14, 16, 18]
else:
SEQUENCES = range(29)
# settings = open(SCRIPT_DIR+"/evaluation/evaluate_tracking.seqmap.val",'w')
# for SEQ in SEQUENCES:
# settings.write("%04d empty 000000 %06d \n"%(SEQ,int(end_frames[SEQ])))
# settings.close()
seqs_frames = [SEQUENCES, [int(end_frames[s]) + 1 for s in SEQUENCES]]
argv = []
tracked = 'car'
argv.append("{}_3d_det_val".format(tracked))
argv.append(Experiment_name)
argv.append('3D')
argv.append(df)
# remove '*' if the trajectories are for each objects are separated in different subfolders
argv.append("CIWT_Pointrcnn" + '*')
argv.append(seqs_frames)
# argv.append(PHASE)
print("start evaluation")
mota, motp, df = run(*argv)
return mota, motp, df
def exp(Experiment_name, hyp_param, df, PHASE, index =-1):
Basline_Name="AB3DMOT"
DATASET_NAME="kitti"
debug_level=3
DATASET=DATASET_DIR+DATASET_NAME+"/{}".format("training")
PREPROC=DATASET_DIR+DATASET_NAME+"/{}/preproc".format("training")
OUT_DIR=DATASET_DIR+DATASET_NAME+"/{}/{}".format("training",Experiment_name)
# Possible modes: 'detection', 'detection_shape'
# The latter is considered obsolete; this option is still here for eval purposes
MODE="detection" # detection_shape
CIWT="/usr/stud/kaddah/ciwt_gd/build/apps/CIWTApp"
# Specify detectors (should correspond to detector dir names and config names)
# DETS=['det_02_regionlets', 'det_02_3DOP', 'pointrcnn']
DETS=['pointrcnn']
end_frames=["153", "446", "232", "143", "313", "296", "269", "799", "389", "802", "293", "372", "77", "339", "105", "375", "208", "144", "338", "1058", "836"]
if DATASET_NAME=="kitti":
if PHASE=="training":
SEQUENCES = [2, 6, 7, 8, 10, 13, 14, 16, 18]
elif PHASE=="validation":
SEQUENCES = [0,1,3,4,5,9,11,12,15,17,19,20]
if not os.path.exists(SCRIPT_DIR+"/results/"+Experiment_name):
if PHASE == "validation":
shutil.copytree(SCRIPT_DIR+"/results/ready_results_baseline_val",SCRIPT_DIR+"/results/"+Experiment_name)
else:
shutil.copytree(SCRIPT_DIR+"/results/ready_results_baseline",SCRIPT_DIR+"/results/"+Experiment_name)
# Exec tracker for each detection set
for DET in DETS:
# Exec tracker for each sequence
for SEQ in SEQUENCES:
SEQNAME="%04d"%(SEQ)
print("Process: {}".format(SEQNAME))
IM_LEFT="{}/image_02/{}/%06d.png".format(DATASET,SEQNAME)
IM_RIGHT="{}/image_03/{}/%06d.png".format(DATASET,SEQNAME)
DISP="{}/disparity/disp_{}/%06d_left.png".format(PREPROC,SEQNAME)
END_FRAME=end_frames[SEQ]
CALIB="{}/calib/{}.txt".format(DATASET,SEQNAME)
PROP="{}/cached_proposals/{}/%06d.bin".format(PREPROC,SEQNAME)
OUT_DIR_DET="{}/{}".format(OUT_DIR,DET)
DETPATH="{}/detection/{}/{}/%06d.txt".format(PREPROC,DET,SEQNAME)
DETCFG="{}/cfg/{}.cfg".format(SCRIPT_DIR,DET)
# if os.path.exists(SCRIPT_DIR+"/results/"+Experiment_name):
# shutil.rmtree(SCRIPT_DIR+"/results/"+Experiment_name, ignore_errors=True)
# shutil.copytree(SCRIPT_DIR+"/results/ready_results_baseline/",SCRIPT_DIR+"/results/"+Experiment_name+"/")
args = ["--config", str(DETCFG),
"--left_image_path", str(IM_LEFT) ,
"--end_frame", str(END_FRAME) ,
"--right_image_path", str(IM_RIGHT),
"--left_disparity_path", str(DISP) ,
"--calib_path", str(CALIB) ,
"--object_proposal_path", str(PROP) ,
"--detections_path", str(DETPATH),
"--output_dir", str(OUT_DIR_DET),
"--tracking_mode", str(MODE),
"--sequence_name", str(SEQNAME),
"--debug_level", str(debug_level) ,
"--is_oriented", str(1) ,
"--enable_coupling", str(1) ,
"--check_exit_to2D_projeciton", str(0) ,
"--association_score_is_3DIoU", str(1)
,
"--association_appearance_model_weight", str(hyp_param["association_appearance_model_weight"]) ,
"--IoU3D_model_weight", str(hyp_param["IoU3D_model_weight"]) ,
"--association_weight_distance_from_camera_param", str(hyp_param["association_weight_distance_from_camera_param"]) ,
"--gaiting_appearance_threshold", str(hyp_param["gaiting_appearance_threshold"]) ,
"--gaiting_IOU_threshold", str(hyp_param["gaiting_IOU_threshold"]) ,
"--gaiting_mh_distance_threshold", str(hyp_param["gaiting_mh_distance_threshold"]) ,
"--gaiting_min_association_score", str(hyp_param["gaiting_min_association_score"]) ,
"--gaiting_size_2D", str(hyp_param["gaiting_size_2D"]) ,
"--tracking_e1", str(hyp_param["tracking_e1"]) ,
"--tracking_e3", str(hyp_param["tracking_e3"]) ,
"--tracking_e4", str(hyp_param["tracking_e4"]) ,
"--id_handling_overlap_threshold", str(hyp_param["id_handling_overlap_threshold"]) ,
"--tracking_exp_decay", str(hyp_param["tracking_exp_decay"]) ,
# "--not_selected_tolerance", str(hyp_param["not_selected_tolerance"]) ,
"--max_hole_size", str(hyp_param["max_hole_size"]) ,
"--tracking_temporal_window_size", str(hyp_param["tracking_temporal_window_size"]) ,
"--hole_penalty_decay_parameter", str(hyp_param["hole_penalty_decay_parameter"]) ,
"--min_observations_needed_to_init_hypothesis", str(hyp_param["min_observations_needed_to_init_hypothesis"]) ,
"--tracking_single_detection_hypo_threshold", str(hyp_param["tracking_single_detection_hypo_threshold"])
# ,
# "--kf_2d_observation_noise", str(hyp["kf_2d_observation_noise"]) ,
# "--kf_2d_initial_variance", str(hyp["kf_2d_initial_variance"]) ,
# "--kf_2d_system_noise", str(hyp["kf_2d_system_noise"]) ,
# "--kf_3d_observation_noise", str(hyp["kf_3d_observation_noise"]) ,
# "--kf_3d_initial_variance",str(hyp["kf_3d_initial_variance"]) ,
# "--kf_3d_system_noise",str(hyp["kf_3d_system_noise"]) ,
# "--kf_system_pos_variance_car_x",str(hyp["kf_system_pos_variance_car_x"]) ,
# "--kf_system_pos_variance_car_z",str(hyp["kf_system_pos_variance_car_z"]) ,
# "--kf_system_velocity_variance_car_x",str(hyp["kf_system_velocity_variance_car_x"]) ,
# "--kf_system_velocity_variance_car_z",str(hyp["kf_system_velocity_variance_car_z"]) ,
# "--kf_orientation_obs_noise",str(hyp["kf_orientation_obs_noise"])
]
subprocess.run([CIWT] + args)
for DET in DETS:
shutil.move(OUT_DIR+"/"+DET+"/data",SCRIPT_DIR+"/results/"+Experiment_name+"/CIWT_Pointrcnn/val/data")
# Determine sequences here
if PHASE =="training":
SEQUENCES = [2, 6, 7, 8, 10, 13, 14, 16, 18]
else:
SEQUENCES = [0,1,3,4,5,9,11,12,15,17,19,20]
seqs_frames = [SEQUENCES, [ int(end_frames[s])+1 for s in SEQUENCES]]
tracked = ['car', 'ped']
DIMENSION = ['3D', '2D']
for dim in DIMENSION:
for obj in tracked:
argv = []
argv.append("{}_3d_det_val".format(obj))
argv.append(Experiment_name)
argv.append(dim)
argv.append(df[dim][obj])
# remove '*' if the trajectories are for each objects are separated in different subfolders
argv.append("CIWT_Pointrcnn"+'*')
argv.append(seqs_frames)
# argv.append(PHASE)
print("start evaluation")
mota, motp, df[dim][obj] = run(*argv)
return df