elif isinstance(first_dp, dict):
result = {}
for key in first_dp.keys():
data_list = [x[key] for x in data_holder]
if use_list:
result[key] = data_list
else:
result[key] = BatchData2Biggest._batch_numpy(data_list)
return result
if __name__ == '__main__':
tensorpack.utils.logger.auto_set_dir()
train_set = MyDataFlow('/media/neil/DATA/mysunrgbd', 'training')
test_set = MyDataFlow('/media/neil/DATA/mysunrgbd',
'training') # TODO: prepare test data
print(
eval_mAP(
sunrgbd_object('/media/neil/DATA/mysunrgbd', 'training'),
OfflinePredictor(
PredictConfig(model=Model(),
input_names=['points'],
output_names=[
'bboxes_pred', 'class_scores_pred',
'batch_idx'
])), [0.25]))
# dataset = BatchData(PrefetchData(train_set, 4, 4), BATCH_SIZE)
for key in first_dp.keys():
data_list = [x[key] for x in data_holder]
if use_list:
result[key] = data_list
else:
result[key] = BatchData2Biggest._batch_numpy(data_list)
return result
if __name__ == '__main__':
tensorpack.utils.logger.auto_set_dir()
# this is the official train/val split
train_set = MyDataFlow('/media/neil/DATA/mysunrgbd',
'training',
idx_list=list(range(5051, 10336)))
# dataset = BatchData(PrefetchData(train_set, 4, 4), BATCH_SIZE)
lr_schedule = [(80, 1e-4), (120, 1e-5)]
# lr_schedule = [(i, 5e-5) for i in range(260)]
# get the config which contains everything necessary in a training
config = AutoResumeTrainConfig(
always_resume=False,
model=Model(),
# The input source for training. FeedInput is slow, this is just for demo purpose.
# In practice it's best to use QueueInput or others. See tutorials for details.
data=QueueInput(
BatchData2Biggest(
PrefetchData(train_set,
elif isinstance(first_dp, dict):
result = {}
for key in first_dp.keys():
data_list = [x[key] for x in data_holder]
if use_list:
result[key] = data_list
else:
result[key] = BatchData2Biggest._batch_numpy(data_list)
return result
if __name__ == '__main__':
logger.auto_set_dir()
# this is the official train/val split
train_set = MyDataFlow('/data/mysunrgbd', 'training', training=True, idx_list=list(range(5051, 10336)), cache_dir='cache_train')
# TestDataSpeed(train_set).start()
gt_cls = {}
gt_all = {}
for classname in type2class:
gt_cls[classname] = get_gt_cls('/home/neil/frustum-pointnets/sunrgbd/sunrgbd_detection/gt_boxes', classname)
for img_id in gt_cls[classname]:
if img_id not in gt_all:
gt_all[img_id] = []
for box in gt_cls[classname][img_id]:
gt_all[img_id].append((classname, box))
# print(classname, box)
lr_schedule = [(80, 1e-4), (120, 1e-5)]
# lr_schedule = [(i, 5e-5) for i in range(260)]