def _linemod_to_coco(cls, split):
data_root = 'data/linemod'
model_path = os.path.join(data_root, cls,
cls + '.ply') # 'data/linemod/cat/cat.ply'
renderer = OpenGLRenderer(model_path)
K = linemod_K #
model = renderer.model['pts'] / 1000 # (94404, 3)
corner_3d = get_model_corners(model) # (8, 3)
center_3d = (np.max(corner_3d, 0) + np.min(corner_3d, 0)) / 2 # (3,)
fps_3d = np.loadtxt(os.path.join(data_root, cls, 'farthest.txt'))
model_meta = {
'K': K,
'corner_3d': corner_3d,
'center_3d': center_3d,
'fps_3d': fps_3d,
'data_root': data_root,
'cls': cls,
'split': split
}
img_id = 0
ann_id = 0
images = []
annotations = []
if split == 'occ':
img_id, ann_id = record_occ_ann(model_meta, img_id, ann_id, images,
annotations)
elif split == 'train' or split == 'test':
img_id, ann_id = record_real_ann(model_meta, img_id, ann_id, images,
annotations)
if split == 'train':
img_id, ann_id = record_fuse_ann(model_meta, img_id, ann_id, images,
annotations)
img_id, ann_id = record_render_ann(model_meta, img_id, ann_id, images,
annotations)
categories = [{'supercategory': 'none', 'id': 1, 'name': cls}]
instance = {
'images': images,
'annotations': annotations,
'categories': categories
}
anno_path = os.path.join(data_root, cls, split + '.json')
with open(anno_path, 'w') as f:
json.dump(instance, f) # 'data/linemod/cat/train.json'
def _tless_train_to_coco(obj_id):
data_root = 'data/tless'
model_path = os.path.join(data_root, 'models_cad',
'obj_{:03}.ply'.format(obj_id))
renderer = OpenGLRenderer(model_path)
model = renderer.model['pts'] / 1000
corner_3d = get_model_corners(model)
center_3d = (np.max(corner_3d, 0) + np.min(corner_3d, 0)) / 2
fps_3d = np.loadtxt(
os.path.join(data_root, 'farthest',
'farthest_{:02}.txt'.format(obj_id)))
model_meta = {
'corner_3d': corner_3d,
'center_3d': center_3d,
'fps_3d': fps_3d,
'data_root': data_root,
'obj_id': obj_id,
}
img_id = 0
ann_id = 0
images = []
annotations = []
img_id, ann_id = record_real_ann(model_meta, img_id, ann_id, images,
annotations)
img_id, ann_id = record_render_ann(model_meta, img_id, ann_id, images,
annotations)
categories = [{'supercategory': 'none', 'id': 1, 'name': obj_id}]
instance = {
'images': images,
'annotations': annotations,
'categories': categories
}
data_cache_dir = 'data/cache/tless_pose/'
obj_cache_dir = os.path.join(data_cache_dir, '{:02}'.format(obj_id))
os.system('mkdir -p {}'.format(obj_cache_dir))
anno_path = os.path.join(obj_cache_dir, 'train.json')
with open(anno_path, 'w') as f:
json.dump(instance, f)
def custom_to_coco(data_root):
model_path = os.path.join(data_root, 'model.ply')
cls_type = input("On which class do you want to train? ")
renderer = OpenGLRenderer(model_path)
K = np.loadtxt(os.path.join(data_root, 'camera.txt'))
model = renderer.model['pts'] / 1000
corner_3d = get_model_corners(model)
print("corner_3d:")
print(corner_3d)
center_3d = (np.max(corner_3d, 0) + np.min(corner_3d, 0)) / 2
print("center_3d:")
print(center_3d)
fps_3d = np.loadtxt(os.path.join(data_root, 'fps.txt'))
model_meta = {
'K': K,
'corner_3d': corner_3d,
'center_3d': center_3d,
'fps_3d': fps_3d,
'data_root': data_root,
}
img_id = 0
ann_id = 0
images = []
annotations = []
img_id, ann_id = record_ann(model_meta, img_id, ann_id, images,
annotations, cls_type)
categories = [{'supercategory': 'none', 'id': 1, 'name': cls_type}]
instance = {
'images': images,
'annotations': annotations,
'categories': categories
}
anno_path = os.path.join(data_root, 'train.json')
with open(anno_path, 'w') as f:
json.dump(instance, f)