])
sp_score = is_dst_pt_inside_dt.mean()
assert (0. <= sp_score <= 1.)
return sp_score
def get_scores(self):
# return self._sift_matches(), self._sift_inliers()
return self._sm(), self._sp()
if __name__ == '__main__':
import pickle
from baselines.sift.test import test
from flowmatch.networks.flownet import FlowNet
from flowmatch.utils import load_config
# Loading an (un-preprocessed) example that was saved as a pickle file.
with open('/home/sancha/repos/osid/sandbox/example.pkl', 'rb') as f:
example = pickle.load(f)
cfg = load_config('../../pipeline/configs/tdid_avd2_manual_easy.yaml')
net = FlowNet(cfg.flownet)
sift_matches = test(net, example)
example['pred_sift'] = sift_matches
# Reclassification.
reclass_scorer = SiftMatchScorer(example)
scores = reclass_scorer.get_scores()
def __len__(self):
return len(self.ids)
def id_to_str(self, id):
return id_to_str(id, block=8)
def str_to_id(self, string):
return str_to_id(string, block=8)
def _init_transform(self):
# Initialize torch-defined transforms that will be used by self.transform().
self.transform_ops = {'ToTensor': transforms.ToTensor()}
def add_gt_flow(self, example):
return example
if __name__ == '__main__':
"""For creating cache"""
parser = argparse.ArgumentParser(
description='Creating an object dataset cache.')
parser.add_argument('--config', default='config', help='config file path')
args = parser.parse_args()
cfg = load_config(args.config)
tg_store = TargetStore(obj_json=cfg.obj_det.obj_json,
obj_root=cfg.obj_det.obj_root,
cfg=cfg)
tg_store.create_cache()
from tqdm import tqdm
from flowmatch.datasets.coco import CocoDataset
from flowmatch.networks.flownet import FlowNet
from flowmatch.flowutils.utils import get_identity_flow
from flowmatch.utils import load_config
from flowmatch.exps.coco.main import coco_filter
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Computing baseline performance on COCO')
parser.add_argument('--config', default='config', help='name of config')
parser.add_argument('--num_examples', '-r', type=int, default=None, help='number of examples to test on')
args = parser.parse_args()
cfg_path = '../../exps/coco/{}.py'.format(args.config)
cfg = load_config(cfg_path)
coco_valid = CocoDataset(root=cfg.coco.valid.image_dir, annFile=cfg.coco.valid.ann_file, cfg=cfg)
print('Filtering training set ... ', end='', flush=True)
filtered_ids = coco_filter(coco_valid)
coco_valid.item_ids = filtered_ids
print('done.')
net = FlowNet(cfg) # just for pre-processing data
# Variables for identity flow
sum_identity_epe = 0.0
identity_flow = get_identity_flow(cfg.tg_size, cfg.tg_size)
# Variables for best constant flow
sum_gt_flow, sum_sq_gt_flow, sum_tg_mask = 0., 0., 0.
res[detections[i]['id']] = detections[i]['score']
else:
next_highest_score = other_scores[over_inds[0]]
res[detections[i]['id']] = max(
0, detections[i]['score'] - next_highest_score)
res = [{'id': x, 'scores': [res[x]] * nvp} for x in res]
return res
def _others(self, arrs, i):
return [np.concatenate([arr[:i], arr[i + 1:]]) for arr in arrs]
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--config',
default='config.yaml',
help='config file path')
parser.add_argument('--eval_ds',
default='gmu_test',
help='dataset to generate SimObj scores')
parser.set_defaults()
args = parser.parse_args()
eval_ds = args.eval_ds
cfg = load_config(args.config).combine
scorer = SimObjScorer(os.path.join(cfg.root, cfg[eval_ds].dt_json))
scorer.create_score_files(os.path.join(cfg.root, cfg[eval_ds].run_dir))