print("Opening", args.base_dir)
sl_npz = np.load(args.base_dir + '/recording.npz')
cloud = sl_npz['events']
idx = sl_npz['index']
discretization = sl_npz['discretization']
K = None
D = None
slice_width = args.nbins * discretization
first_ts = cloud[0][0]
last_ts = cloud[-1][0]
vis_dir = os.path.join(args.base_dir, 'vis')
pydvs.replace_dir(vis_dir)
print("The recording range:", first_ts, "-", last_ts)
print("Discretization resolution:", discretization)
import matplotlib.pyplot as plt
ebins = []
for i in range(len(idx) - args.nbins):
nevents = idx[i + args.nbins] - idx[i]
ebins.append(nevents)
median = np.median(ebins)
cutoff = [median for b in ebins]
for i, b in enumerate(ebins):
lo = max(0, i - args.nbins * 10)
hi = min(len(ebins), i + args.nbins * 10)
def data_creator(sequence: str):
print(f"{pydvs.okb('STARTING DATASET CREATION')} for sequence: {sequence}")
dataset_txt = eval(open(os.path.join(sequence, 'meta.txt')).read())
K = np.array([[0.0, 0.0, 0.0], [0.0, 0.0, 0.0], [0.0, 0.0, 1.0]])
D = np.array([0.0, 0.0, 0.0, 0.0])
K[0][0] = dataset_txt['meta']['fx']
K[1][1] = dataset_txt['meta']['fy']
K[0][2] = dataset_txt['meta']['cx']
K[1][2] = dataset_txt['meta']['cy']
D[0] = dataset_txt['meta']['k1']
D[1] = dataset_txt['meta']['k2']
D[2] = dataset_txt['meta']['k3']
D[3] = dataset_txt['meta']['k4']
RES_X = dataset_txt['meta']['res_x']
RES_Y = dataset_txt['meta']['res_y']
NUM_FRAMES = len(dataset_txt['frames'])
frames_meta = dataset_txt['frames']
oids = []
for key in frames_meta[0]:
if (key == 'cam'): continue
if (type(frames_meta[0][key]) == type(dict())
and 'pos' in frames_meta[0][key]):
oids.append(key)
print(pydvs.okb("Resolution:"), RES_X, 'x', RES_Y)
print(pydvs.okb("Frames:"), NUM_FRAMES)
print(pydvs.okb("Object ids:"), oids)
print(pydvs.okb("Calibration:"))
print(K)
print(D)
# Create a plot
#save_plot(frames_meta, oids, os.path.join(args.base_dir, 'position_plots.pdf'), tp='pos')
#save_plot(dataset_txt['full_trajectory'], oids, os.path.join(args.base_dir, 'position_plots.pdf'), tp='pos')
# Read depth / masks
print(pydvs.bld("Reading the depth and masks:"))
depths = np.zeros((NUM_FRAMES, ) + (RES_X, RES_Y), dtype=np.uint16)
masks = np.zeros((NUM_FRAMES, ) + (RES_X, RES_Y), dtype=np.uint16)
classical = np.zeros((NUM_FRAMES, ) + (RES_X, RES_Y, 3), dtype=np.uint8)
classical_read = 0
for i, frame in enumerate(frames_meta):
print("frame\t", i + 1, "/", NUM_FRAMES, "\t", end='\r')
gt_frame_name = os.path.join(sequence, f"img/{frame['gt_frame']}")
gt_img = cv2.imread(gt_frame_name, cv2.IMREAD_UNCHANGED)
if (gt_img.dtype != depths.dtype or gt_img.dtype != masks.dtype):
print("\tType mismatch! Expected", depths.dtype, " but have",
gt_img.dtype)
sys.exit(-1)
depths[i, :, :] = gt_img[:, :, 0] # depth is in mm
masks[i, :, :] = gt_img[:, :, 2] # mask is object ids * 1000
if ('classical_frame' in frame.keys()):
classical_frame_name = os.path.join(
sequence, f"img/{frame['classical_frame']}")
classical_img = cv2.imread(classical_frame_name,
cv2.IMREAD_UNCHANGED)
classical[i, :, :, :] = classical_img
if (gt_img.dtype != depths.dtype or gt_img.dtype != masks.dtype):
print("\tType mismatch! Expected", classical.dtype,
" but have", classical_img.dtype)
sys.exit(-1)
classical_read += 1
print("\n")
if (classical_read > 0):
print(pydvs.okb("Read "), classical_read, "/", NUM_FRAMES,
pydvs.okb(" classical frames"))
else:
classical = None
# Read event cloud
cloud, idx = pydvs.read_event_file_txt(
os.path.join(sequence, 'events.txt'), args.discretization)
tmin = frames_meta[0]['ts']
tmax = frames_meta[-1]['ts']
if (cloud.shape[0] > 0):
tmin = cloud[0][0]
tmax = cloud[-1][0]
print(pydvs.okb("The recording range:"), tmin, "-", tmax)
print(pydvs.okb("The gt range:"), frames_meta[0]['ts'], "-",
frames_meta[-1]['ts'])
print(pydvs.okb("Discretization resolution:"), args.discretization)
"""
# Save .npz file
print (pydvs.bld("Saving..."))
np.savez_compressed(os.path.join(args.base_dir, 'dataset.npz'), events=cloud, index=idx, classical=classical,
discretization=args.discretization, K=K, D=D, depth=depths, mask=masks, meta=dataset_txt)
print ("\n")
"""
# Generate images:
slice_dir = os.path.join(sequence, 'slices')
#vis_dir = os.path.join(args.base_dir, 'vis')
pydvs.replace_dir(slice_dir)
#pydvs.replace_dir(vis_dir)
for i, frame in enumerate(frames_meta):
print("Saving sanity check frames\t",
i + 1,
"/",
NUM_FRAMES,
"\t",
end='\r')
time = frame['ts']
if (time > tmax or time < tmin):
continue
cv2.imwrite(
os.path.join(slice_dir, 'mask_' + str(i).rjust(10, '0') + '.png'),
masks[i].astype(np.uint16))
if (cloud.shape[0] > 0):
sl, _ = pydvs.get_slice(cloud, idx, time, args.slice_width, 1,
args.discretization)
eimg = dvs_img(sl, (RES_X, RES_Y),
None,
None,
args.slice_width,
mode=1)
cv2.imwrite(
os.path.join(slice_dir,
'frame_' + str(i).rjust(10, '0') + '.png'), eimg)
depth = depths[i].astype(np.float)
mask = masks[i].astype(np.float)
col_mask = mask_to_color(mask)
"""
# normalize for visualization
mask = (255 * (mask.astype(np.float) - np.nanmin(mask)) / (np.nanmax(mask) - np.nanmin(mask))).astype(np.uint8)
depth = (255 * (depth.astype(np.float) - np.nanmin(depth)) / (np.nanmax(depth) - np.nanmin(depth))).astype(np.uint8)
if ((classical_read > 0) and (classical is not None)):
grayscale_img = cv2.cvtColor(classical[i], cv2.COLOR_BGR2GRAY).astype(np.float)
rgb_img = np.dstack((grayscale_img, grayscale_img, grayscale_img))
rgb_img[mask > 0] = rgb_img[mask > 0] * 0.2 + col_mask[mask > 0] * 0.8
#rgb_img = np.rot90(rgb_img, k=2)
#depth = np.rot90(depth, k=2)
eimg = np.hstack((rgb_img.astype(np.uint8), np.dstack((depth,depth,depth))))
else:
eimg = dvs_img(sl, (RES_X, RES_Y), None, None, args.slice_width, mode=0)
eimg[mask > 0] = eimg[mask > 0] * 0.5 + col_mask[mask > 0] * 0.5
eimg = np.hstack((eimg.astype(np.uint8), np.dstack((depth,depth,depth))))
#footer = gen_text_stub(eimg.shape[1], frame)
#eimg = np.vstack((eimg, footer))
cv2.imwrite(os.path.join(vis_dir, 'frame_' + str(i).rjust(10, '0') + '.png'), eimg)
"""
print(F"{pydvs.okg('FINISHED DATASET CREATION')} for sequence {sequence}")