def test_handle():
# handle=Handle_Rawjpg()
# handle = Handle_Npyarr()
# test
#-# handle=Handle_Rawjpg()
#-# imgID = 'n02690373_16'
#-# imgpath = '/Users/shine/working/cvpr17pose/dataset/PASCAL3D/ImageData.Max500/aeroplane_imagenet/n02690373_16.jpg'
handle = Handle_Rawpng()
imgID = '1507795919477055'
imgpath = '/Users/shine/Pictures/1507795919477055.png'
#
k, v = handle.pack(imgID, imgpath)
#
cv2.imshow('image-pack', handle.unpack(v))
cv2_wait()
img_arr = cv2.imread(imgpath)
k, v = handle.pack_from_npyarr(imgID, img_arr)
#
cv2.imshow('image-pack_from_npyarr', handle.unpack(v))
cv2_wait()
print(len(v))
exit()
def _vis_minibatch(self, sample_batched):
"""Visualize a mini-batch for debugging."""
for i, (idx, label, a, e, t, data) in enumerate(
zip(
sample_batched['idx'], # note: these are tensors
sample_batched['label'],
sample_batched['a'],
sample_batched['e'],
sample_batched['t'],
sample_batched['data'])):
rcobj = self.recs[idx]
print(idx, rcobj.obj_id)
im = data.numpy().transpose((1, 2, 0)).copy()
im = (im * self.roiloader_pascal3d.pxl_std
) + self.roiloader_pascal3d.pxl_mean
if self.mode == 'torchmodel':
im = (im * 255)[:, :, ::-1].astype(np.uint8) # RBG->BGR
else: # caffemodel type
im = im.astype(np.uint8)
text = '%s %.1f' % (rcobj.category, a)
cv2_putText(im, (0, 20), text, bgcolor=(255, 255, 255)) #
text = ' a=%.1f,e=%.1f,t=%.1f' % (rcobj.gt_view.a, rcobj.gt_view.e,
rcobj.gt_view.t)
cv2_putText(im, (0, 40), text, bgcolor=(255, 255, 255)) #
cv2.imshow('im', im)
cv2_wait()
def test_read_db1():
imgdb = ImageData_lmdb(
env.Home +
'/working/cvpr17pose/dataset/PASCAL3D/ImageData.Max500.Rawjpg.lmdb')
print(imgdb.keys[:10])
for imgID in imgdb.keys:
img = imgdb[imgID]
cv2.imshow('img', img)
cv2_wait()
def test(visualize=True):
datadb = ImageData_lmdb(db_path)
print (datadb.keys)
for k in datadb.keys:
img = datadb[k]
if visualize: # :
cv2.imshow('real_img', img)
cv2_wait()
def test_img_lmdb():
imgdb = ImageData_lmdb(
env.Home +
'/working/cvpr17pose/dataset/PASCAL3D/ImageData.Max500.Rawjpg.lmdb')
#imgID = np.string_('n03693474_19496') # np.string_
#imgID = np.unicode_('n03693474_19496') # np.string_
#imgID = np.string_('n03693474_19496').tobytes() # np.bytes_ ?
imgID = 'n03693474_19496' # str
#imgID = b'n03693474_19496' # bytes
im = imgdb[imgID]
cv2.imshow('image', im)
cv2_wait()
def test_lmdb():
# lmdb_env = lmdb.open(os.path.join(PASCAL3D_Dir,'ImageData.Max500.Rawjpg.lmdb'), map_size=map_size)
lmdb_env = lmdb.open(
env.Home +
'/working/cvpr17pose/dataset/PASCAL3D/ImageData.Max500.Rawjpg.lmdb',
map_size=10000)
handle = Handle_Rawjpg()
imgID = b'n03693474_19496'
with lmdb_env.begin() as txn:
raw_data = txn.get(imgID) #.tobytes()) #(b'00000000')
cv2.imshow('image', handle.unpack(raw_data))
cv2_wait()
def vis(self,
nr_row=5,
nr_col=6,
side=150,
randshow=False,
is_depth_img=False): # nr_per_row=6,
# Iter by keys
keys = self.keys
if randshow:
random.shuffle(keys)
itkeys = iter(keys)
if True:
# with self.lmdb_env.begin() as txn:
row_images = []
rows = []
# for key, raw_data in txn.cursor():
# assert raw_data is not None
# pad_img = pad_as_squared_img(self.handle.unpack(raw_data), Side=side)
# Iter by keys
for key in itkeys:
next_img = self[key]
if is_depth_img: # [TODO] TO REMOVE this case.
_max, _min = next_img.max(), next_img.min()
next_img = (next_img.astype(np.float32) - _min) / (
_max - _min) # normalize [min,max] to [0,1]
elif self.db_path.endswith(
'.Rawpng.lmdb') and self.handle.remap is not None:
_max, _min = self.handle.remap['max'], self.handle.remap[
'min']
next_img = (next_img - _min) / (_max - _min)
pad_img = pad_as_squared_img(next_img, Side=side)
if len(row_images) < nr_col:
row_images.append(pad_img)
else:
cat_imgs = np.concatenate(row_images, axis=1)
row_images = [pad_img] # reset row_images
if len(rows) < nr_row:
rows.append(cat_imgs)
else:
cat_alls = np.concatenate(rows, axis=0)
cv2.imshow('images', cat_alls)
cv2_wait()
rows = [cat_imgs] # reset rows
def _vis_minibatch(self, sample_batched):
"""Visualize a mini-batch for debugging."""
for i, (idx, label, quat, data) in enumerate(
zip(
sample_batched['idx'], # note: these are tensors
sample_batched['label'],
sample_batched['quat'],
sample_batched['data'])):
rc = self.recs[idx]
# print idx
im = data.numpy().transpose((1, 2, 0)).copy()
im += self.cfg.PIXEL_MEANS
im = im.astype(np.uint8) # xmin, ymax
a, b, c, d = quat
cv2_putText(im, (0, 20), rc.category, bgcolor=(255, 255, 255))
text = '%.1f %.1f %.1f %.1f' % (a, b, c, d)
cv2_putText(im, (0, 40), text, bgcolor=(255, 255, 255))
cv2.imshow('im', im)
cv2_wait()
def _vis_one(self, idx, data, norm, mask, flip):
data = data.numpy() if isinstance(data, torch.Tensor) else data
norm = norm.numpy() if isinstance(norm, torch.Tensor) else norm
mask = mask.numpy() if isinstance(mask, torch.Tensor) else mask
flip = flip.numpy() if isinstance(flip, torch.Tensor) else flip
#
_norm = norm.reshape(-1, 3)
_mask = mask.astype(np.bool).reshape(-1)
_norm = _norm[_mask]
#
if True:
sampling = 0.01
_inds = np.arange(len(_norm))
sample_inds = np.random.choice(_inds,
size=int(len(_inds) * sampling),
replace=False)
_nodes = _norm[sample_inds]
nodes = np.zeros_like(_nodes)
nodes[:, 0], nodes[:,
1], nodes[:,
2] = _nodes[:,
0], _nodes[:,
2], _nodes[:,
1]
visualize3d(nodes)
#
im = data.transpose((1, 2, 0)).copy()
im += caffe_mean_pxl # mean_pxl
im = im.astype(np.uint8) # xmin, ymax
#
norm = ((norm + 1) * (2**7)).astype(
np.uint8) # map from [-1,1] to [0,256]
#
mask = cv2.cvtColor(mask, cv2.COLOR_GRAY2BGR)
#
showim = np.concatenate([im, norm, mask], axis=1)
cv2_putText(showim, (0, 20), "with_flip: %s " % flip.astype(np.bool))
cv2.imshow('im', showim)
cv2_wait()
