def build_model_from_cfg(config_path, checkpoint_path, cfg_options=None):
"""Build a model from config and load the given checkpoint.
Args:
config_path (str): the OpenMMLab config for the model we want to
export to ONNX
checkpoint_path (str): Path to the corresponding checkpoint
Returns:
torch.nn.Module: the built model
"""
from mmdet.models import build_detector
cfg = mmcv.Config.fromfile(config_path)
if cfg_options is not None:
cfg.merge_from_dict(cfg_options)
# set cudnn_benchmark
if cfg.get('cudnn_benchmark', False):
torch.backends.cudnn.benchmark = True
cfg.model.pretrained = None
cfg.data.test.test_mode = True
# build the model
cfg.model.train_cfg = None
model = build_detector(cfg.model, test_cfg=cfg.get('test_cfg'))
checkpoint = load_checkpoint(model, checkpoint_path, map_location='cpu')
if 'CLASSES' in checkpoint.get('meta', {}):
model.CLASSES = checkpoint['meta']['CLASSES']
else:
from mmdet.datasets import DATASETS
dataset = DATASETS.get(cfg.data.test['type'])
assert (dataset is not None)
model.CLASSES = dataset.CLASSES
model.cpu().eval()
return model
def _dist_train(model, dataset, cfg, validate=False):
# prepare data loaders
try:
selsa_imgs = cfg.data.selsa_imgs
except Exception:
selsa_imgs = cfg.data.imgs_per_gpu
dataset = dataset if isinstance(dataset, (list, tuple)) else [dataset]
data_loaders = [
build_dataloader(ds,
cfg.data.imgs_per_gpu,
cfg.data.workers_per_gpu,
selsa_imgs=selsa_imgs,
dist=True) for ds in dataset
]
# put model on gpus
model = MMDistributedDataParallel(model.cuda())
mod = cfg.pop('mod', False)
# build runner
optimizer = build_optimizer(model, cfg.optimizer, mod)
runner = Runner(model, batch_processor, optimizer, cfg.work_dir,
cfg.log_level)
# fp16 setting
fp16_cfg = cfg.get('fp16', None)
if fp16_cfg is not None:
optimizer_config = Fp16OptimizerHook(**cfg.optimizer_config,
**fp16_cfg)
else:
optimizer_config = DistOptimizerHook(**cfg.optimizer_config)
# register hooks
runner.register_training_hooks(cfg.lr_config, optimizer_config,
cfg.checkpoint_config, cfg.log_config)
runner.register_hook(DistSamplerSeedHook())
# register eval hooks
if validate:
val_dataset_cfg = cfg.data.val
eval_cfg = cfg.get('evaluation', {})
if isinstance(model.module, RPN):
# TODO: implement recall hooks for other datasets
runner.register_hook(
CocoDistEvalRecallHook(val_dataset_cfg, **eval_cfg))
else:
dataset_type = DATASETS.get(val_dataset_cfg.type)
if issubclass(dataset_type, datasets.CocoDataset):
runner.register_hook(
CocoDistEvalmAPHook(val_dataset_cfg, **eval_cfg))
else:
runner.register_hook(
DistEvalmAPHook(val_dataset_cfg, **eval_cfg))
if cfg.resume_from:
runner.resume(cfg.resume_from)
elif cfg.load_from:
runner.load_checkpoint(cfg.load_from)
runner.run(data_loaders, cfg.workflow, cfg.total_epochs)
def _dist_train(model,
dataset,
cfg,
validate=False,
logger=None,
timestamp=None):
# prepare data loaders
dataset = dataset if isinstance(dataset, (list, tuple)) else [dataset]
data_loaders = [
build_dataloader(
ds, cfg.data.imgs_per_gpu, cfg.data.workers_per_gpu, dist=True)
for ds in dataset
]
# put model on gpus
model = MMDistributedDataParallel(model.cuda())
# build runner
optimizer = build_optimizer(model, cfg.optimizer)
runner = Runner(
model, batch_processor, optimizer, cfg.work_dir, logger=logger)
# an ugly walkaround to make the .log and .log.json filenames the same
runner.timestamp = timestamp
# fp16 setting
fp16_cfg = cfg.get('fp16', None)
if fp16_cfg is not None:
optimizer_config = Fp16OptimizerHook(**cfg.optimizer_config,
**fp16_cfg)
else:
optimizer_config = DistOptimizerHook(**cfg.optimizer_config)
# register hooks
runner.register_training_hooks(cfg.lr_config, optimizer_config,
cfg.checkpoint_config, cfg.log_config)
runner.register_hook(DistSamplerSeedHook())
# register eval hooks
if validate:
val_dataset_cfg = cfg.data.val
eval_cfg = cfg.get('evaluation', {})
if isinstance(model.module, RPN):
# TODO: implement recall hooks for other datasets
runner.register_hook(
CocoDistEvalRecallHook(val_dataset_cfg, **eval_cfg))
else:
dataset_type = DATASETS.get(val_dataset_cfg.type)
if issubclass(dataset_type, datasets.CocoDataset) or issubclass(dataset_type, datasets.Acoustic):
runner.register_hook(
CocoDistEvalmAPHook(val_dataset_cfg, **eval_cfg))
else:
runner.register_hook(
DistEvalmAPHook(val_dataset_cfg, **eval_cfg))
if cfg.resume_from:
runner.resume(cfg.resume_from)
elif cfg.load_from:
runner.load_checkpoint(cfg.load_from)
runner.run(data_loaders, cfg.workflow, cfg.total_epochs)
def _non_dist_train(model,
dataset,
cfg,
validate=False,
logger=None,
timestamp=None):
# prepare data loaders
dataset = dataset if isinstance(dataset, (list, tuple)) else [dataset]
data_loaders = [
build_dataloader(ds,
cfg.data.imgs_per_gpu,
cfg.data.workers_per_gpu,
cfg.gpus,
dist=False) for ds in dataset
]
# put model on gpus
model = MMDataParallel(model, device_ids=range(cfg.gpus)).cuda()
# build runner
optimizer = build_optimizer(model, cfg.optimizer)
runner = Runner(model,
batch_processor,
optimizer,
cfg.work_dir,
logger=logger)
# an ugly walkaround to make the .log and .log.json filenames the same
runner.timestamp = timestamp
# fp16 setting
fp16_cfg = cfg.get('fp16', None)
if fp16_cfg is not None:
optimizer_config = Fp16OptimizerHook(**cfg.optimizer_config,
**fp16_cfg,
distributed=False)
else:
optimizer_config = cfg.optimizer_config
runner.register_training_hooks(cfg.lr_config, optimizer_config,
cfg.checkpoint_config, cfg.log_config)
if validate:
val_dataset_cfg = cfg.data.val
eval_cfg = cfg.get('evaluation', {})
dataset_type = DATASETS.get(val_dataset_cfg.type)
if issubclass(dataset_type, datasets.CocoDataset):
runner.register_hook(
CocoLocalEvalmAPHook(val_dataset_cfg, **eval_cfg))
if cfg.resume_from:
runner.resume(cfg.resume_from)
elif cfg.load_from:
runner.load_checkpoint(cfg.load_from)
import warnings
with warnings.catch_warnings():
warnings.simplefilter("ignore")
runner.run(data_loaders, cfg.workflow, cfg.total_epochs)
def test_xml_dataset():
dataconfig = {
'ann_file': 'data/VOCdevkit/VOC2007/ImageSets/Main/test.txt',
'img_prefix': 'data/VOCdevkit/VOC2007/',
'pipeline': [{
'type': 'LoadImageFromFile'
}]
}
XMLDataset = DATASETS.get('XMLDataset')
class XMLDatasetSubClass(XMLDataset):
CLASSES = None
# get_ann_info and _filter_imgs of XMLDataset
# would use self.CLASSES, we added CLASSES not NONE
with pytest.raises(AssertionError):
XMLDatasetSubClass(**dataconfig)
def test_allow_empty_images(classes, expected_length):
dataset_class = DATASETS.get('CocoDataset')
# Filter empty images
filtered_dataset = dataset_class(ann_file='tests/data/coco_sample.json',
img_prefix='tests/data',
pipeline=[],
classes=classes,
filter_empty_gt=True)
# Get all
full_dataset = dataset_class(ann_file='tests/data/coco_sample.json',
img_prefix='tests/data',
pipeline=[],
classes=classes,
filter_empty_gt=False)
assert len(filtered_dataset) == expected_length
assert len(filtered_dataset.img_ids) == expected_length
assert len(full_dataset) == 3
assert len(full_dataset.img_ids) == 3
assert filtered_dataset.CLASSES == classes
assert full_dataset.CLASSES == classes
def test_custom_classes_override_default(dataset):
dataset_class = DATASETS.get(dataset)
if dataset in ['CocoDataset', 'CityscapesDataset']:
dataset_class.coco = MagicMock()
dataset_class.cat_ids = MagicMock()
original_classes = dataset_class.CLASSES
# Test setting classes as a tuple
custom_dataset = dataset_class(
ann_file=MagicMock(),
pipeline=[],
classes=('bus', 'car'),
test_mode=True,
img_prefix='VOC2007' if dataset == 'VOCDataset' else '')
assert custom_dataset.CLASSES != original_classes
assert custom_dataset.CLASSES == ('bus', 'car')
# Test setting classes as a list
custom_dataset = dataset_class(
ann_file=MagicMock(),
pipeline=[],
classes=['bus', 'car'],
test_mode=True,
img_prefix='VOC2007' if dataset == 'VOCDataset' else '')
assert custom_dataset.CLASSES != original_classes
assert custom_dataset.CLASSES == ['bus', 'car']
# Test overriding not a subset
custom_dataset = dataset_class(
ann_file=MagicMock(),
pipeline=[],
classes=['foo'],
test_mode=True,
img_prefix='VOC2007' if dataset == 'VOCDataset' else '')
assert custom_dataset.CLASSES != original_classes
assert custom_dataset.CLASSES == ['foo']
# Test default behavior
custom_dataset = dataset_class(
ann_file=MagicMock(),
pipeline=[],
classes=None,
test_mode=True,
img_prefix='VOC2007' if dataset == 'VOCDataset' else '')
assert custom_dataset.CLASSES == original_classes
# Test sending file path
import tempfile
tmp_file = tempfile.NamedTemporaryFile()
with open(tmp_file.name, 'w') as f:
f.write('bus\ncar\n')
custom_dataset = dataset_class(
ann_file=MagicMock(),
pipeline=[],
classes=tmp_file.name,
test_mode=True,
img_prefix='VOC2007' if dataset == 'VOCDataset' else '')
tmp_file.close()
assert custom_dataset.CLASSES != original_classes
assert custom_dataset.CLASSES == ['bus', 'car']
def main():
args = parse_args()
assert args.out or args.show or args.json_out, \
('Please specify at least one operation (save or show the results) '
'with the argument "--out" or "--show" or "--json_out"')
if args.out is not None and not args.out.endswith(('.pkl', '.pickle')):
raise ValueError('The output file must be a pkl file.')
if args.json_out is not None and args.json_out.endswith('.json'):
args.json_out = args.json_out[:-5]
cfg = mmcv.Config.fromfile(args.config)
# set cudnn_benchmark
if cfg.get('cudnn_benchmark', False):
torch.backends.cudnn.benchmark = True
cfg.model.pretrained = None
cfg.data.test.test_mode = True
# init distributed env first, since logger depends on the dist info.
if args.launcher == 'none':
distributed = False
else:
distributed = True
init_dist(args.launcher, **cfg.dist_params)
# build the dataloader
# TODO: support multiple images per gpu (only minor changes are needed)
dataset = build_dataset(cfg.data.test)
data_loader = build_dataloader(dataset,
imgs_per_gpu=1,
workers_per_gpu=cfg.data.workers_per_gpu,
dist=distributed,
shuffle=False)
# build the model and load checkpoint
model = build_detector(cfg.model, train_cfg=None, test_cfg=cfg.test_cfg)
fp16_cfg = cfg.get('fp16', None)
if fp16_cfg is not None:
wrap_fp16_model(model)
checkpoint = load_checkpoint(model, args.checkpoint, map_location='cpu')
# old versions did not save class info in checkpoints, this walkaround is
# for backward compatibility
if 'CLASSES' in checkpoint['meta']:
model.CLASSES = checkpoint['meta']['CLASSES']
else:
model.CLASSES = dataset.CLASSES
if not distributed:
model = MMDataParallel(model, device_ids=[0])
outputs = single_gpu_test(model, data_loader, args.show)
else:
model = MMDistributedDataParallel(model.cuda())
outputs = multi_gpu_test(model, data_loader, args.tmpdir)
rank, _ = get_dist_info()
if args.out and rank == 0:
print('\nwriting results to {}'.format(args.out))
mmcv.dump(outputs, args.out)
eval_types = args.eval
if eval_types:
print('Starting evaluate {}'.format(' and '.join(eval_types)))
if eval_types == ['proposal_fast']:
result_file = args.out
coco_eval(result_file, eval_types, dataset.coco)
else:
if not isinstance(outputs[0], dict):
result_files = results2json(dataset, outputs, args.out)
dataset_type = DATASETS.get(cfg.data.test.type)
if issubclass(dataset_type, datasets.CocoDataset):
coco_eval(result_files, eval_types, dataset.coco)
elif issubclass(dataset_type, datasets.LVISDataset):
max_dets = cfg.test_cfg['rcnn']['max_per_img']
lvis_eval(result_files,
eval_types,
dataset.lvis,
max_dets=max_dets)
else:
raise ValueError(
'{} is not supported type for evaluation'.format(
dataset_type))
else:
for name in outputs[0]:
print('\nEvaluating {}'.format(name))
outputs_ = [out[name] for out in outputs]
result_file = args.out + '.{}'.format(name)
result_files = results2json(dataset, outputs_,
result_file)
coco_eval(result_files, eval_types, dataset.coco)
# Save predictions in the COCO json format
if args.json_out and rank == 0:
if not isinstance(outputs[0], dict):
results2json(dataset, outputs, args.json_out)
else:
for name in outputs[0]:
outputs_ = [out[name] for out in outputs]
result_file = args.json_out + '.{}'.format(name)
results2json(dataset, outputs_, result_file)
def setUp(self):
super().setUp()
self.data_dir = osp.join(
osp.dirname(osp.dirname(osp.dirname(__file__))), 'data')
self.dataset_class = DATASETS.get('XMLDataset')
def test_custom_classes_override_default(dataset):
dataset_class = DATASETS.get(dataset)
if dataset in ['CocoDataset', 'CityscapesDataset']:
dataset_class.coco = MagicMock()
dataset_class.cat_ids = MagicMock()
original_classes = dataset_class.CLASSES
# Test setting classes as a tuple
custom_dataset = dataset_class(
ann_file=MagicMock(),
pipeline=[],
classes=('bus', 'car'),
test_mode=True,
img_prefix='VOC2007' if dataset == 'VOCDataset' else '')
assert custom_dataset.CLASSES != original_classes
assert custom_dataset.CLASSES == ('bus', 'car')
print(custom_dataset)
# Test setting classes as a list
custom_dataset = dataset_class(
ann_file=MagicMock(),
pipeline=[],
classes=['bus', 'car'],
test_mode=True,
img_prefix='VOC2007' if dataset == 'VOCDataset' else '')
assert custom_dataset.CLASSES != original_classes
assert custom_dataset.CLASSES == ['bus', 'car']
print(custom_dataset)
# Test overriding not a subset
custom_dataset = dataset_class(
ann_file=MagicMock(),
pipeline=[],
classes=['foo'],
test_mode=True,
img_prefix='VOC2007' if dataset == 'VOCDataset' else '')
assert custom_dataset.CLASSES != original_classes
assert custom_dataset.CLASSES == ['foo']
print(custom_dataset)
# Test default behavior
custom_dataset = dataset_class(
ann_file=MagicMock(),
pipeline=[],
classes=None,
test_mode=True,
img_prefix='VOC2007' if dataset == 'VOCDataset' else '')
assert custom_dataset.CLASSES == original_classes
print(custom_dataset)
# Test sending file path
import tempfile
with tempfile.TemporaryDirectory() as tmpdir:
path = tmpdir + 'classes.txt'
with open(path, 'w') as f:
f.write('bus\ncar\n')
custom_dataset = dataset_class(
ann_file=MagicMock(),
pipeline=[],
classes=path,
test_mode=True,
img_prefix='VOC2007' if dataset == 'VOCDataset' else '')
assert custom_dataset.CLASSES != original_classes
assert custom_dataset.CLASSES == ['bus', 'car']
print(custom_dataset)
def _dist_train(model, dataset, cfg, validate=False):
# prepare data loaders
dataset = dataset if isinstance(dataset, (list, tuple)) else [dataset]
if 'use_img_sampling' not in cfg.data:
cfg.data.update({'use_img_sampling': False})
if 'use_sample_out' not in cfg.data:
cfg.data.update({'use_sample_out': False})
print('--Dist-train--IS:{}--ISout:{}'.format(cfg.data.use_img_sampling,
cfg.data.use_sample_out))
data_loaders = [
build_dataloader(ds,
cfg.data.imgs_per_gpu,
cfg.data.workers_per_gpu,
dist=True,
use_img_sampling=cfg.data.use_img_sampling,
use_sample_out=cfg.data.use_sample_out)
for ds in dataset
]
# put model on gpus
model = MMDistributedDataParallel(model.cuda())
# build runner
optimizer = build_optimizer(model, cfg.optimizer)
runner = Runner(model, batch_processor, optimizer, cfg.work_dir,
cfg.log_level)
# Add for LVIS by LiYu
import logging
runner.logger.setLevel(logging.INFO)
# ====================
# fp16 setting
fp16_cfg = cfg.get('fp16', None)
if fp16_cfg is not None:
optimizer_config = Fp16OptimizerHook(**cfg.optimizer_config,
**fp16_cfg)
else:
optimizer_config = DistOptimizerHook(**cfg.optimizer_config)
# register hooks
runner.register_training_hooks(cfg.lr_config, optimizer_config,
cfg.checkpoint_config, cfg.log_config)
runner.register_hook(DistSamplerSeedHook())
# register eval hooks
if validate:
val_dataset_cfg = cfg.data.val
eval_cfg = cfg.get('evaluation', {})
if isinstance(model.module, RPN):
# TODO: implement recall hooks for other datasets
runner.register_hook(
CocoDistEvalRecallHook(val_dataset_cfg, **eval_cfg))
else:
dataset_type = DATASETS.get(val_dataset_cfg.type)
if issubclass(dataset_type, datasets.CocoDataset):
runner.register_hook(
CocoDistEvalmAPHook(val_dataset_cfg, **eval_cfg))
else:
runner.register_hook(
DistEvalmAPHook(val_dataset_cfg, **eval_cfg))
if cfg.resume_from:
runner.resume(cfg.resume_from)
elif cfg.load_from:
runner.load_checkpoint(cfg.load_from)
runner.run(data_loaders, cfg.workflow, cfg.total_epochs)
def _dist_train(model,
dataset,
cfg,
validate=False,
logger=None,
timestamp=None):
# prepare data loaders
dataset = dataset if isinstance(dataset, (list, tuple)) else [dataset]
data_loaders = [
build_dataloader(
ds, cfg.data.imgs_per_gpu, cfg.data.workers_per_gpu, dist=True)
for ds in dataset
]
# put model on gpus
model = MMDistributedDataParallel(model.cuda())
# # if model.module.bbox_head.freeze_solov2_and_train_combonly:
# if model.module.bbox_head.optimize_list is not None:
# for (key, param) in model.named_parameters():
# # if 'kernel_convs_convcomb' not in key and 'context_fusion_convs' not in key and 'learned_weight' not in key:
# if not any(s in key for s in model.module.bbox_head.optimize_list):
# param.requires_grad=False
# else:
# # print('optimize {}'.format(key))
# logger.info('optimize {}'.format(key))
# build runner
optimizer = build_optimizer(model, cfg.optimizer)
runner = Runner(
model, batch_processor, optimizer, cfg.work_dir, logger=logger)
# an ugly walkaround to make the .log and .log.json filenames the same
runner.timestamp = timestamp
# fp16 setting
fp16_cfg = cfg.get('fp16', None)
if fp16_cfg is not None:
optimizer_config = Fp16OptimizerHook(**cfg.optimizer_config,
**fp16_cfg)
else:
optimizer_config = DistOptimizerHook(**cfg.optimizer_config)
# register hooks
runner.register_training_hooks(cfg.lr_config, optimizer_config,
cfg.checkpoint_config, cfg.log_config)
runner.register_hook(DistSamplerSeedHook())
# register eval hooks
if validate:
val_dataset_cfg = cfg.data.val
eval_cfg = cfg.get('evaluation', {})
if isinstance(model.module, RPN):
# TODO: implement recall hooks for other datasets
runner.register_hook(
CocoDistEvalRecallHook(val_dataset_cfg, **eval_cfg))
else:
dataset_type = DATASETS.get(val_dataset_cfg.type)
if issubclass(dataset_type, datasets.CocoDataset):
runner.register_hook(
CocoDistEvalmAPHook(val_dataset_cfg, **eval_cfg))
else:
runner.register_hook(
DistEvalmAPHook(val_dataset_cfg, **eval_cfg))
if cfg.resume_from:
runner.resume(cfg.resume_from)
elif cfg.load_from:
runner.load_checkpoint(cfg.load_from)
runner.run(data_loaders, cfg.workflow, cfg.total_epochs)
## add test after training
if cfg.data.test.ann_file != 'data/lvis/lvis_v0.5_val_lvis_freqset.json': # if val set is lvis freq, only eval on lvis-freq val set
cfg.data.test.test_mode = True
dataset = build_dataset(cfg.data.test)
data_loader = build_dataloader(
dataset,
imgs_per_gpu=1,
workers_per_gpu=cfg.data.workers_per_gpu,
dist=False,
shuffle=False)
model_orig=model.module
model = MMDataParallel(model, device_ids=[0]).cuda()
# data_loader.dataset.img_infos = data_loader.dataset.img_infos[:100]
outputs = single_gpu_test(model, data_loader)
print('\nwriting results to {}'.format('xxx'))
# mmcv.dump(outputs, 'xxx')
eval_types = ['segm']
if eval_types:
print('Starting evaluate {}'.format(' and '.join(eval_types)))
if eval_types == ['proposal_fast']:
result_file = 'xxx'
coco_eval(result_file, eval_types, dataset.coco)
else:
if not isinstance(outputs[0], dict):
result_files = results2json_segm(dataset, outputs, 'xxx', dump=False)
coco_eval(result_files, eval_types, dataset.coco)
else:
for name in outputs[0]:
print('\nEvaluating {}'.format(name))
outputs_ = [out[name] for out in outputs]
result_file = 'xxx' + '.{}'.format(name)
result_files = results2json(dataset, outputs_,
result_file, dump=False)
coco_eval(result_files, eval_types, dataset.coco)
##eval on lvis-77######
cfg.data.test.ann_file = 'data/lvis/lvis_v0.5_val_cocofied.json'
cfg.data.test.img_prefix = 'data/lvis/val2017/'
cfg.data.test.test_mode = True
dataset = build_dataset(cfg.data.test)
data_loader = build_dataloader(
dataset,
imgs_per_gpu=1,
workers_per_gpu=cfg.data.workers_per_gpu,
dist=False,
shuffle=False)
# model_orig=model.module
# model = MMDataParallel(model, device_ids=[0]).cuda()
# data_loader.dataset.img_infos = data_loader.dataset.img_infos[:100]
outputs = single_gpu_test(model, data_loader)
print('\nwriting results to {}'.format('xxx'))
# mmcv.dump(outputs, 'xxx')
eval_types = ['segm']
if eval_types:
print('Starting evaluate {}'.format(' and '.join(eval_types)))
if eval_types == ['proposal_fast']:
result_file = 'xxx'
coco_eval(result_file, eval_types, dataset.coco)
else:
if not isinstance(outputs[0], dict):
result_files = results2json_segm(dataset, outputs, 'xxx', dump=False)
from lvis import LVISEval
lvisEval = LVISEval('data/lvis/lvis_v0.5_val_cocofied.json', result_files, 'segm')
lvisEval.run()
lvisEval.print_results()
# fix lvis api eval iou_thr error, should be 0.9 but was 0.8999
lvisEval.params.iou_thrs[8] = 0.9
for iou in [0.5, 0.6, 0.7, 0.8, 0.9]:
print('AP at iou {}: {}'.format(iou, lvisEval._summarize('ap', iou_thr=iou)))
else:
for name in outputs[0]:
print('\nEvaluating {}'.format(name))
outputs_ = [out[name] for out in outputs]
result_file = 'xxx' + '.{}'.format(name)
result_files = results2json(dataset, outputs_,
result_file, dump=False)
coco_eval(result_files, eval_types, dataset.coco)
else:
##eval on lvis-freq######
cfg.data.test.test_mode = True
dataset = build_dataset(cfg.data.test)
data_loader = build_dataloader(
dataset,
imgs_per_gpu=1,
workers_per_gpu=cfg.data.workers_per_gpu,
dist=False,
shuffle=False)
# model_orig=model.module
# model = MMDataParallel(model, device_ids=[0]).cuda()
data_loader.dataset.img_infos = data_loader.dataset.img_infos[:100]
outputs = single_gpu_test(model, data_loader)
print('\nwriting results to {}'.format('xxx'))
# mmcv.dump(outputs, 'xxx')
eval_types = ['segm']
if eval_types:
print('Starting evaluate {}'.format(' and '.join(eval_types)))
if eval_types == ['proposal_fast']:
result_file = 'xxx'
coco_eval(result_file, eval_types, dataset.coco)
else:
if not isinstance(outputs[0], dict):
result_files = results2json_segm(dataset, outputs, 'xxx', dump=False)
from lvis import LVISEval
lvisEval = LVISEval(cfg.data.test.ann_file, result_files, 'segm')
lvisEval.run()
lvisEval.print_results()
# fix lvis api eval iou_thr error, should be 0.9 but was 0.8999
lvisEval.params.iou_thrs[8] = 0.9
for iou in [0.5, 0.6, 0.7, 0.8, 0.9]:
print('AP at iou {}: {}'.format(iou, lvisEval._summarize('ap', iou_thr=iou)))
else:
for name in outputs[0]:
print('\nEvaluating {}'.format(name))
outputs_ = [out[name] for out in outputs]
result_file = 'xxx' + '.{}'.format(name)
result_files = results2json(dataset, outputs_,
result_file, dump=False)
coco_eval(result_files, eval_types, dataset.coco)
input_shape = parse_shape(args.shape)
else:
img_scale = cfg.test_pipeline[1]['img_scale']
input_shape = (1, 3, img_scale[1], img_scale[0])
if not args.max_shape:
max_shape = input_shape
else:
max_shape = parse_shape(args.max_shape)
if not args.min_shape:
min_shape = input_shape
else:
min_shape = parse_shape(args.min_shape)
dataset = DATASETS.get(cfg.data.test['type'])
assert (dataset is not None)
CLASSES = dataset.CLASSES
normalize_cfg = parse_normalize_cfg(cfg.test_pipeline)
input_config = {
'min_shape': min_shape,
'opt_shape': input_shape,
'max_shape': max_shape,
'input_shape': input_shape,
'input_path': args.input_img,
'normalize_cfg': normalize_cfg
}
# Create TensorRT engine
onnx2tensorrt(args.model,
args.trt_file,
def _dist_train(model,
dataset,
cfg,
validate=False,
logger=None,
timestamp=None):
# prepare data loaders
dataset = dataset if isinstance(dataset, (list, tuple)) else [dataset]
data_loaders = [
build_dataloader(
ds, cfg.data.imgs_per_gpu, cfg.data.workers_per_gpu, dist=True)
for ds in dataset
]
# for index, item in enumerate(data_loaders):
# a = iter(item)
# print('~~~', index, a, next(a))
# break
from mmdet.apis import get_root_logger
logger = get_root_logger()
logger.info(model)
model = MMDistributedDataParallel(model.cuda())
# build runner
optimizer = build_optimizer(model, cfg.optimizer)
runner = Runner(
model, batch_processor, optimizer, cfg.work_dir, logger=logger)
# an ugly walkaround to make the .log and .log.json filenames the same
runner.timestamp = timestamp
# fp16 setting
fp16_cfg = cfg.get('fp16', None)
if fp16_cfg is not None:
optimizer_config = Fp16OptimizerHook(**cfg.optimizer_config,
**fp16_cfg)
else:
optimizer_config = DistOptimizerHook(**cfg.optimizer_config)
# register hooks
runner.register_training_hooks(cfg.lr_config, optimizer_config,
cfg.checkpoint_config, cfg.log_config, cfg.total_epochs, len(data_loaders[0]))
runner.register_hook(DistSamplerSeedHook())
# register eval hooks
if validate:
val_dataset_cfg = cfg.data.val
eval_cfg = cfg.get('evaluation', {})
if isinstance(model.module, RPN):
# TODO: implement recall hooks for other datasets
runner.register_hook(
CocoDistEvalRecallHook(val_dataset_cfg, **eval_cfg))
else:
dataset_type = DATASETS.get(val_dataset_cfg.type)
if cfg.data.val.type in ['KittiDataset', 'KittiInCocoDataset', 'Kitti3dDataset']:
runner.register_hook(
KITTIDistEvalmAPHook(val_dataset_cfg, **eval_cfg))
elif issubclass(dataset_type, datasets.CocoDataset):
logger.info('Using CocoDistEvalmAPHook.')
runner.register_hook(
CocoDistEvalmAPHook(val_dataset_cfg, **eval_cfg))
else:
runner.register_hook(
DistEvalmAPHook(val_dataset_cfg, **eval_cfg))
if cfg.resume_from:
runner.resume(cfg.resume_from)
elif cfg.load_from:
runner.load_checkpoint(cfg.load_from)
runner.run(data_loaders, cfg.workflow, cfg.total_epochs)
def _dist_train(model,
dataset,
cfg,
validate=False,
logger=None,
timestamp=None,
runner_attr_dict=dict()):
# prepare data loaders
dataset = dataset if isinstance(dataset, (list, tuple)) else [dataset]
data_loaders = [
build_dataloader(ds,
cfg.data.imgs_per_gpu,
cfg.data.workers_per_gpu,
dist=True) for ds in dataset
]
# build runner
runner_attr_dict.update({
'imgs_per_gpu': cfg.data.imgs_per_gpu,
'initial_lr': cfg.optimizer['lr']
})
if hasattr(dataset, 'CLASSES'):
runner_attr_dict.update({'classes': dataset.CLASSES})
optimizer = build_optimizer(model, cfg.optimizer)
search_optimizer = getattr(cfg, 'search_config',
{}).pop('search_optimizer', None)
runner = Runner(model,
batch_processor,
optimizer,
search_optimizer,
cfg.work_dir,
logger=logger,
runner_attr_dict=runner_attr_dict)
# an ugly walkaround to make the .log and .log.json filenames the same
runner.timestamp = timestamp
# fp16 setting
fp16_cfg = cfg.get('fp16', None)
if fp16_cfg is not None:
optimizer_config = Fp16OptimizerHook(**cfg.optimizer_config,
**fp16_cfg)
else:
optimizer_config = DistOptimizerHook(**cfg.optimizer_config)
# register hooks
runner.register_training_hooks(cfg.lr_config, optimizer_config)
runner.register_hook(DistSamplerSeedHook())
if search_optimizer is not None:
runner.register_hook(DistSearchOptimizerHook())
runner.register_hook(SearchHook(**cfg.search_config))
# register eval hooks
if validate:
val_dataset_cfg = cfg.data.val
eval_cfg = cfg.get('evaluation', {})
if isinstance(model.module, RPN):
# TODO: implement recall hooks for other datasets
runner.register_hook(
CocoDistEvalRecallHook(val_dataset_cfg, **eval_cfg))
else:
dataset_type = DATASETS.get(val_dataset_cfg.type)
if issubclass(dataset_type, datasets.CocoDataset):
runner.register_hook(
CocoDistEvalmAPHook(val_dataset_cfg, **eval_cfg))
else:
runner.register_hook(
DistEvalmAPHook(val_dataset_cfg, **eval_cfg))
register_hooks(runner, cfg)
if cfg.resume_from:
runner.resume(cfg.resume_from)
elif cfg.load_from:
runner.load_checkpoint(cfg.load_from)
runner.run(data_loaders, cfg.workflow, cfg.total_epochs)