def _run_test_cat(num_inputs, layout, ndim, axis, axis_name):
num_iter = 3
batch_size = 4
if ndim is None:
ndim = len(layout)
ref_axis = layout.find(axis_name) if axis_name is not None else axis if axis is not None else 0
assert ref_axis >= 0
axis_arg = None if axis_name else axis
pipe = dali.pipeline.Pipeline(batch_size=batch_size, num_threads=3, device_id=0)
with pipe:
inputs = fn.external_source(
input_generator(num_inputs, batch_size, ndim, ref_axis),
num_outputs=num_inputs, layout=layout)
out_cpu = fn.cat(*inputs, axis=axis_arg, axis_name=axis_name)
out_gpu = fn.cat(*(x.gpu() for x in inputs), axis=axis_arg, axis_name=axis_name)
pipe.set_outputs(out_cpu, out_gpu, *inputs)
pipe.build()
for iter in range(num_iter):
o_cpu, o_gpu, *inputs = pipe.run()
ref = ref_cat(inputs, ref_axis)
check_batch(o_cpu, ref, batch_size, eps=0, expected_layout=layout)
check_batch(o_gpu, ref, batch_size, eps=0, expected_layout=layout)
def test_cat_empty_input():
pipe = dali.pipeline.Pipeline(batch_size = 1, num_threads = 3, device_id = 0)
with pipe:
src1 = dali.types.Constant(np.array(
[[1, 2, 3 ,4],
[5, 6, 7, 8],
[9,10,11,12]]))
src2 = dali.types.Constant(np.array(
[[],
[],
[]], dtype=np.int32))
src3 = dali.types.Constant(np.array(
[[13,14,15],
[16,17,18],
[19,20,21]]))
out_cpu = fn.cat(src1, src2, src3, axis = 1)
out_gpu = fn.cat(src1.gpu(), src2.gpu(), src3.gpu(), axis = 1)
pipe.set_outputs(out_cpu, out_gpu)
pipe.build()
o = pipe.run()
o = list(o)
o[1] = o[1].as_cpu();
ref = np.array([[1, 2, 3, 4,13,14,15],
[5, 6, 7, 8,16,17,18],
[9,10,11,12,19,20,21]])
assert np.array_equal(o[0].at(0), ref)
assert np.array_equal(o[1].at(0), ref)
def define_graph(self):
inputs, bboxes, labels, polygons, vertices = fn.readers.coco(
file_root=self.file_root,
annotations_file=self.annotation_file,
skip_empty=True,
shard_id=self.share_id,
num_shards=self.num_gpus,
ratio=True,
ltrb=True,
polygon_masks = True,
random_shuffle=self.random_shuffle,
shuffle_after_epoch=self.shuffle_after_epoch,
name="Reader")
input_shape = fn.slice(fn.cast(fn.peek_image_shape(inputs), dtype=types.INT32), 0, 2, axes=[0])
h = fn.slice(input_shape, 0, 1, axes = [0], dtype=types.FLOAT)
w = fn.slice(input_shape, 1, 1, axes = [0], dtype=types.FLOAT)
short_side = math.min(w, h)
scale = fn.random.uniform(range=[0.3, 1.])
crop_side = fn.cast(math.ceil(scale * short_side), dtype=types.INT32)
crop_shape = fn.cat(crop_side, crop_side)
anchor_rel, shape_rel, bboxes, labels, bbox_indices = fn.random_bbox_crop(
bboxes,
labels,
input_shape=input_shape,
crop_shape=crop_shape,
shape_layout="HW",
thresholds=[0.], # No minimum intersection-over-union, for demo purposes
allow_no_crop=False, # No-crop is disallowed, for demo purposes
seed=-1, # Fixed random seed for deterministic results
bbox_layout="xyXY", # left, top, right, back
output_bbox_indices=True, # Output indices of the filtered bounding boxes
total_num_attempts=1024,
)
polygons, vertices = fn.segmentation.select_masks(
bbox_indices, polygons, vertices
)
images = fn.decoders.image_slice(
inputs, anchor_rel, shape_rel, normalized_anchor=False, normalized_shape=False, device='mixed'
)
images = fn.color_space_conversion(images, image_type=types.RGB, output_type=types.BGR)
MT_1_vertices = fn.transforms.crop(
to_start=(0.0, 0.0), to_end=fn.cat(w, h)
)
MT_2_vertices = fn.transforms.crop(
from_start=anchor_rel, from_end=(anchor_rel + shape_rel),
to_start=(0.0, 0.0), to_end=(1., 1.)
)
vertices = fn.coord_transform(fn.coord_transform(vertices, MT=MT_1_vertices), MT=MT_2_vertices)
targets = fn.cat( bboxes, fn.reshape(vertices, shape=[-1, 10]), axis=1)
interp_methods = [types.INTERP_LINEAR, types.INTERP_CUBIC, types.INTERP_LANCZOS3, types.INTERP_GAUSSIAN, types.INTERP_NN, types.INTERP_TRIANGULAR]
interp_method = fn.random.uniform(values=[int(x) for x in interp_methods], dtype=types.INT32)
interp_method = fn.reinterpret(interp_method, dtype=types.INTERP_TYPE)
images = fn.resize(images, dtype=types.FLOAT, size=self.input_dim, interp_type=interp_method)
labels = labels.gpu()
targets = targets.gpu()
return (images, targets, labels)
def test_cat_numpy_array():
pipe = dali.pipeline.Pipeline(1,1,0)
src = fn.external_source([[np.array([[10,11],[12,13]], dtype=np.float32)]])
pipe.set_outputs(fn.cat(src, np.array([[20],[21]], dtype=np.float32), axis=1))
pipe.build()
o = pipe.run()
assert np.array_equal(o[0].at(0), np.array([[10,11,20],[12,13,21]]))
def incorrect_input_sets_pipeline():
jpegs, _ = fn.readers.file(file_root=file_root,
seed=42,
random_shuffle=True)
images = fn.decoders.image(jpegs, seed=42)
output = fn.cat([images, images, images], [images, images])
return tuple(output)
def kwargs_len_change():
input = [np.zeros(1)] * 8
inputs = [input] * 2
kwargs = {}
if kwargs_len_change.change:
kwargs_len_change.change = False
kwargs['axis'] = 0
return fn.cat(*inputs, **kwargs)
def inputs_len_change():
input = [np.zeros(1)] * 8
if inputs_len_change.change:
inputs_len_change.change = False
inputs = [input]
else:
inputs = [input] * 2
return fn.cat(*inputs)
def multiple_input_sets_pipeline():
jpegs = [
fn.readers.file(file_root=file_root, seed=42, random_shuffle=True)[0]
for _ in range(6)
]
images = fn.decoders.image(jpegs, seed=42)
cropped_images = fn.random_resized_crop(images, size=(224, 224), seed=42)
output = fn.cat(cropped_images[:3], cropped_images[3:])
return tuple(output)
def test_cat_cpu():
pipe = Pipeline(batch_size=batch_size, num_threads=3, device_id=None)
data = fn.external_source(source=get_data, layout="HWC")
data2 = fn.external_source(source=get_data, layout="HWC")
data3 = fn.external_source(source=get_data, layout="HWC")
pixel_pos = fn.cat(data, data2, data3)
pipe.set_outputs(pixel_pos)
pipe.build()
for _ in range(3):
pipe.run()
def test_cat_all_empty():
pipe = dali.pipeline.Pipeline(batch_size = 1, num_threads = 3, device_id = 0)
with pipe:
src1 = dali.types.Constant(np.array(
[[],
[],
[]], dtype=np.int32))
out_cpu = fn.cat(src1, src1, src1, axis = 1)
out_gpu = fn.cat(src1.gpu(), src1.gpu(), src1.gpu(), axis = 1)
pipe.set_outputs(out_cpu, out_gpu)
pipe.build()
o = pipe.run()
o = list(o)
o[1] = o[1].as_cpu();
ref = np.array([[], [], []], dtype=np.int32)
assert np.array_equal(o[0].at(0), ref)
assert np.array_equal(o[1].at(0), ref)
def dali_frame_splicing_graph(x, nfeatures, x_len, stacking=1, subsampling=1):
if stacking > 1:
seq = [x]
for n in range(1, stacking):
f = fn.slice(x,
n,
x_len,
axes=(1, ),
out_of_bounds_policy='pad',
fill_values=0)
seq.append(f)
x = fn.cat(*seq, axis=0)
nfeatures = nfeatures * stacking
if subsampling > 1:
out_len = (x_len + subsampling - 1) // subsampling
m = fn.transforms.scale(scale=[subsampling, 1], center=[0.5, 0])
x = fn.reshape(x, rel_shape=[1, 1, -1],
layout="HWC") # Layout required by WarpAffine
size = fn.cat(nfeatures, out_len)
x = fn.warp_affine(x, matrix=m, size=size, interp_type=types.INTERP_NN)
x = fn.reshape(x, rel_shape=[1, 1], layout="ft")
return x
def tfrecord_pipeline(dspath, batch_size, num_threads, device="cpu", device_id=None,
shard_id=0, num_shards=1, reader_name="Reader",
seq=True, chroms=False, chroms_vlog=False, target=True, target_vlog=True, label=False, random_shuffle=True):
pipe = Pipeline(batch_size=batch_size, num_threads=num_threads, device_id=device_id)
feature_description = {}
feature_description["seq"] = tfrec.VarLenFeature(tfrec.float32, -1.0)
feature_description["label"] = tfrec.FixedLenFeature([], tfrec.int64, -1)
feature_description["target"] = tfrec.FixedLenFeature([], tfrec.float32, -1.0)
for ct in dspath["chromatin_tracks"]:
feature_description[ct] = tfrec.VarLenFeature(tfrec.float32, -1.0)
with pipe:
inputs = fn.readers.tfrecord(
name=reader_name,
path=dspath['TFRecord'],
index_path=dspath['TFRecord_idx'],
features=feature_description,
shard_id = shard_id,
num_shards = num_shards,
random_shuffle=random_shuffle,
read_ahead=True,
prefetch_queue_depth=20,
pad_last_batch=True)
if device=="gpu":
inputs['seq'] = inputs['seq'].gpu()
for ct in dspath["chromatin_tracks"]: inputs[ct] = inputs[ct].gpu()
inputs['target'] = inputs['target'].gpu()
inputs['label'] = inputs['label'].gpu()
seqdata = fn.expand_dims(inputs['seq'], axes=1, device=device)
seqdata = fn.reshape(seqdata, shape=(4, -1), device=device)
chromsdata = fn.cat(*[fn.expand_dims(inputs[ct], axes=0, device=device) for ct in dspath["chromatin_tracks"]], axis=0, device=device)
sample = []
if seq: sample.append(seqdata)
if chroms:
if chroms_vlog:
sample.append(log(chromsdata + 1))
else:
sample.append(chromsdata)
if target:
if target_vlog:
sample.append(log(inputs['target'] + 1))
else:
sample.append(inputs['target'])
if label: sample.append(inputs['label'])
pipe.set_outputs(*sample)
return pipe
def dali_reflect_pad_graph(x, x_len, pad_amount):
def flip_1d(x):
# TODO(janton): remove the layout trick when Flip supports arbitrary data layouts
x = fn.reshape(x, shape=(-1, 1, 1), layout="HWC")
x = fn.flip(x, vertical=1)
x = fn.reshape(x, shape=(-1, ), layout="t")
return x
pad_start = fn.slice(x, 1, pad_amount, axes=(0, ))
pad_start = flip_1d(pad_start)
pad_end = fn.slice(x, x_len - pad_amount - 1, pad_amount, axes=(0, ))
pad_end = flip_1d(pad_end)
x = fn.cat(pad_start, x, pad_end, axis=0)
return x
def check_random_mask_pixel(ndim=2, batch_size=3,
min_extent=20, max_extent=50):
pipe = dali.pipeline.Pipeline(batch_size=batch_size, num_threads=4, device_id=0, seed=1234)
with pipe:
# Input mask
in_shape_dims = [fn.cast(fn.random.uniform(range=(min_extent, max_extent + 1), shape=(1,), device='cpu'),
dtype=types.INT32) for d in range(ndim)]
in_shape = fn.cat(*in_shape_dims, axis=0)
in_mask = fn.cast(fn.random.uniform(range=(0, 2), device='cpu', shape=in_shape), dtype=types.INT32)
fg_pixel1 = fn.segmentation.random_mask_pixel(in_mask, foreground=1) # > 0
fg_pixel2 = fn.segmentation.random_mask_pixel(in_mask, foreground=1, threshold=0.99) # > 0.99
fg_pixel3 = fn.segmentation.random_mask_pixel(in_mask, foreground=1, value=2) # == 2
rnd_pixel = fn.segmentation.random_mask_pixel(in_mask, foreground=0)
coin_flip = fn.random.coin_flip(probability=0.7)
fg_biased = fn.segmentation.random_mask_pixel(in_mask, foreground=coin_flip)
# Demo purposes: Taking a random pixel and produce a valid anchor to feed slice
crop_shape = in_shape - 2 # We want to force the center adjustment, therefore the large crop shape
anchor = fn.cast(fg_pixel1, dtype=types.INT32) - crop_shape // 2
anchor = math.min(math.max(0, anchor), in_shape - crop_shape)
out_mask = fn.slice(in_mask, anchor, crop_shape, axes=tuple(range(ndim)))
pipe.set_outputs(in_mask, fg_pixel1, fg_pixel2, fg_pixel3, rnd_pixel, coin_flip, fg_biased,
anchor, crop_shape, out_mask)
pipe.build()
for iter in range(3):
outputs = pipe.run()
for idx in range(batch_size):
in_mask = outputs[0].at(idx)
fg_pixel1 = outputs[1].at(idx).tolist()
fg_pixel2 = outputs[2].at(idx).tolist()
fg_pixel3 = outputs[3].at(idx).tolist()
rnd_pixel = outputs[4].at(idx).tolist()
coin_flip = outputs[5].at(idx).tolist()
fg_biased = outputs[6].at(idx).tolist()
anchor = outputs[7].at(idx).tolist()
crop_shape = outputs[8].at(idx).tolist()
out_mask = outputs[9].at(idx)
assert in_mask[tuple(fg_pixel1)] > 0
assert in_mask[tuple(fg_pixel2)] > 0.99
assert in_mask[tuple(fg_pixel3)] == 2
assert in_mask[tuple(fg_biased)] > 0 or not coin_flip
for d in range(ndim):
assert 0 <= anchor[d] and anchor[d] + crop_shape[d] <= in_mask.shape[d]
assert out_mask.shape == tuple(crop_shape)
def check_pad_to_square(device='cpu', batch_size=3, ndim=2, num_iter=3):
pipe = Pipeline(batch_size=batch_size,
num_threads=3,
device_id=0,
seed=1234)
axes = (0, 1)
with pipe:
in_shape = fn.cast(fn.random.uniform(range=(10, 20), shape=(ndim, )),
dtype=types.INT32)
in_data = fn.reshape(fn.random.uniform(range=(0., 1.), shape=in_shape),
layout="HW")
shape = fn.shapes(in_data, dtype=types.INT32)
h = fn.slice(shape, 0, 1, axes=[0])
w = fn.slice(shape, 1, 1, axes=[0])
side = math.max(h, w)
if device == 'gpu':
in_data = in_data.gpu()
out_data = fn.pad(in_data,
axis_names="HW",
shape=fn.cat(side, side, axis=0))
pipe.set_outputs(in_data, out_data)
pipe.build()
for _ in range(num_iter):
outs = [
out.as_cpu() if isinstance(out, TensorListGPU) else out
for out in pipe.run()
]
for i in range(batch_size):
in_data, out_data = \
[outs[out_idx].at(i) for out_idx in range(len(outs))]
in_shape = in_data.shape
max_side = max(in_shape)
for s in out_data.shape:
assert s == max_side
np.testing.assert_equal(out_data[:in_shape[0], :in_shape[1]],
in_data)
np.testing.assert_equal(out_data[in_shape[0]:, :], 0)
np.testing.assert_equal(out_data[:, in_shape[1]:], 0)
def check_bbox_random_crop_adjust_polygons(file_root,
annotations_file,
batch_size=3,
num_iters=4,
num_threads=4,
device_id=0,
seed=1234):
pipe = Pipeline(batch_size=batch_size,
num_threads=num_threads,
device_id=device_id,
seed=seed)
with pipe:
# Read data from COCO
# ratio=True means both bboxes and masks coordinates will be
# relative to the image dimensions (range [0.0, 1.0])
inputs, in_bboxes, labels, in_polygons, in_vertices = \
fn.readers.coco(
file_root=file_root, annotations_file=annotations_file, shard_id=0, num_shards=1,
ratio=True, ltrb=True, polygon_masks=True
)
# Generate a random crop. out_bboxes are adjusted to the crop window
slice_anchor, slice_shape, out_bboxes, labels, bbox_indices = \
fn.random_bbox_crop(
in_bboxes, labels,
aspect_ratio=[0.5, 2.0], thresholds=[0, 0.1, 0.3, 0.5, 0.7, 0.9],
scaling=[0.3, 1.0], bbox_layout='xyXY', output_bbox_indices=True
)
# Crop the image
_ = fn.decoders.image_slice(inputs,
slice_anchor,
slice_shape,
device='mixed',
axis_names='WH')
sel_polygons, sel_vertices = fn.segmentation.select_masks(
bbox_indices, in_polygons, in_vertices)
# Adjust masks coordinates to the coordinate space of the cropped image
MT = fn.transforms.crop(from_start=slice_anchor,
from_end=(slice_anchor + slice_shape))
out_vertices = fn.coord_transform(sel_vertices, MT=MT)
# Converting to absolute coordinates (demo purposes)
image_shape = fn.peek_image_shape(inputs, dtype=types.FLOAT)
h = fn.slice(image_shape, 0, 1, axes=[0])
w = fn.slice(image_shape, 1, 1, axes=[0])
# Original bboxes
bbox_x = fn.slice(in_bboxes, 0, 1, axes=[1])
bbox_y = fn.slice(in_bboxes, 1, 1, axes=[1])
bbox_X = fn.slice(in_bboxes, 2, 1, axes=[1])
bbox_Y = fn.slice(in_bboxes, 3, 1, axes=[1])
in_bboxes_abs = fn.cat(bbox_x * w,
bbox_y * h,
bbox_X * w,
bbox_Y * h,
axis=1)
# Transform to convert relative coordinates to absolute
scale_rel_to_abs = fn.transforms.scale(scale=fn.cat(w, h))
# Selected vertices (relative coordinates)
sel_vertices_abs = fn.coord_transform(out_vertices,
MT=scale_rel_to_abs)
# Output bboxes
bbox2_x = fn.slice(out_bboxes, 0, 1, axes=[1])
bbox2_y = fn.slice(out_bboxes, 1, 1, axes=[1])
bbox2_X = fn.slice(out_bboxes, 2, 1, axes=[1])
bbox2_Y = fn.slice(out_bboxes, 3, 1, axes=[1])
out_bboxes_abs = fn.cat(bbox2_x * w,
bbox2_y * h,
bbox2_X * w,
bbox2_Y * h,
axis=1)
# Output vertices (absolute coordinates)
out_vertices_abs = fn.coord_transform(out_vertices,
MT=scale_rel_to_abs)
# Clamped coordinates
out_vertices_clamped = math.clamp(out_vertices, 0.0, 1.0)
out_vertices_clamped_abs = fn.coord_transform(out_vertices_clamped,
MT=scale_rel_to_abs)
pipe.set_outputs(in_vertices, sel_vertices, sel_vertices_abs, out_vertices,
out_vertices_clamped, out_vertices_abs,
out_vertices_clamped_abs, in_bboxes, in_bboxes_abs,
out_bboxes, out_bboxes_abs, in_polygons, sel_polygons,
image_shape, slice_anchor, slice_shape, bbox_indices)
pipe.build()
# Enough iterations to see an example with more than one bounding box
for i in range(num_iters):
outs = pipe.run()
for j in range(batch_size):
(in_vertices, sel_vertices, sel_vertices_abs, out_vertices,
out_vertices_clamped, out_vertices_abs, out_vertices_clamped_abs,
in_bboxes, in_bboxes_abs, out_bboxes, out_bboxes_abs, in_polygons,
sel_polygons, image_shape, slice_anchor, slice_shape,
bbox_indices) = (outs[k].at(j) for k in range(len(outs)))
# Checking that the output polygon descriptors are the ones associated with the
# selected bounding boxes
expected_polygons_list = []
expected_vertices_list = []
ver_count = 0
for k in range(in_polygons.shape[0]):
mask_id = in_polygons[k][0]
in_ver_start_idx = in_polygons[k][1]
in_ver_end_idx = in_polygons[k][2]
pol_nver = in_ver_end_idx - in_ver_start_idx
if mask_id in bbox_indices:
expected_polygons_list.append(
[mask_id, ver_count, ver_count + pol_nver])
for j in range(in_ver_start_idx, in_ver_end_idx):
expected_vertices_list.append(in_vertices[j])
ver_count = ver_count + pol_nver
expected_sel_polygons = np.array(expected_polygons_list)
np.testing.assert_equal(expected_sel_polygons, sel_polygons)
# Checking the selected vertices correspond to the selected masks
expected_sel_vertices = np.array(expected_vertices_list)
np.testing.assert_equal(expected_sel_vertices, sel_vertices)
# Chekc that the vertices are correctly mapped to the cropping window
expected_out_vertices = np.copy(expected_sel_vertices)
crop_x, crop_y = slice_anchor
crop_w, crop_h = slice_shape
for v in range(expected_out_vertices.shape[0]):
expected_out_vertices[v, 0] = (expected_out_vertices[v, 0] -
crop_x) / crop_w
expected_out_vertices[v, 1] = (expected_out_vertices[v, 1] -
crop_y) / crop_h
np.testing.assert_allclose(expected_out_vertices,
out_vertices,
rtol=1e-4)
# Checking the conversion to absolute coordinates
h, w, _ = image_shape
wh = np.array([w, h])
whwh = np.array([w, h, w, h])
expected_out_vertices_abs = expected_out_vertices * wh
np.testing.assert_allclose(expected_out_vertices_abs,
out_vertices_abs,
rtol=1e-4)
# Checking clamping of the relative coordinates
expected_out_vertices_clamped = np.clip(expected_out_vertices,
a_min=0.0,
a_max=1.0)
np.testing.assert_allclose(expected_out_vertices_clamped,
out_vertices_clamped,
rtol=1e-4)
# Checking clamping of the absolute coordinates
expected_out_vertices_clamped_abs = np.clip(
expected_out_vertices_abs, 0, wh)
np.testing.assert_allclose(expected_out_vertices_clamped_abs,
out_vertices_clamped_abs,
rtol=1e-4)
# Checking scaling of the bounding boxes
expected_in_bboxes_abs = in_bboxes * whwh
np.testing.assert_allclose(expected_in_bboxes_abs,
in_bboxes_abs,
rtol=1e-4)
# Check box selection and mapping to the cropping window
expected_out_bboxes = np.copy(in_bboxes[bbox_indices, :])
for k in range(expected_out_bboxes.shape[0]):
expected_out_bboxes[k, 0] = (expected_out_bboxes[k, 0] -
crop_x) / crop_w
expected_out_bboxes[k, 1] = (expected_out_bboxes[k, 1] -
crop_y) / crop_h
expected_out_bboxes[k, 2] = (expected_out_bboxes[k, 2] -
crop_x) / crop_w
expected_out_bboxes[k, 3] = (expected_out_bboxes[k, 3] -
crop_y) / crop_h
expected_out_bboxes = np.clip(expected_out_bboxes,
a_min=0.0,
a_max=1.0)
np.testing.assert_allclose(expected_out_bboxes,
out_bboxes,
rtol=1e-4)
expected_out_bboxes_abs = expected_out_bboxes * whwh
np.testing.assert_allclose(expected_out_bboxes_abs,
out_bboxes_abs,
rtol=1e-4)
def create_image_pipeline(
batch_size,
num_threads,
device_id,
image0_list,
image1_list,
flow_list,
valBool,
):
pipeline = Pipeline(batch_size, num_threads, device_id, seed=2)
with pipeline:
if valBool:
shuffleBool = False
else:
shuffleBool = True
""" READ FILES """
image0, _ = fn.readers.file(
file_root=args.data,
files=image0_list,
random_shuffle=shuffleBool,
name="Reader",
seed=1,
)
image1, _ = fn.readers.file(
file_root=args.data,
files=image1_list,
random_shuffle=shuffleBool,
seed=1,
)
flo = fn.readers.numpy(
file_root=args.data,
files=flow_list,
random_shuffle=shuffleBool,
seed=1,
)
""" DECODE AND RESHAPE """
image0 = fn.decoders.image(image0, device="cpu")
image0 = fn.reshape(image0, layout="HWC")
image1 = fn.decoders.image(image1, device="cpu")
image1 = fn.reshape(image1, layout="HWC")
images = fn.cat(image0, image1, axis=2)
flo = fn.reshape(flo, layout="HWC")
if valBool:
images = fn.resize(images, resize_x=162, resize_y=122)
else:
""" CO-TRANSFORM """
# random translate
# angle_rng = fn.random.uniform(range=(-90, 90))
# images = fn.rotate(images, angle=angle_rng, fill_value=0)
# flo = fn.rotate(flo, angle=angle_rng, fill_value=0)
images = fn.random_resized_crop(
images,
size=[122, 162], # 122, 162
random_aspect_ratio=[1.3, 1.4],
random_area=[0.8, 0.9],
seed=1,
)
flo = fn.random_resized_crop(
flo,
size=[122, 162],
random_aspect_ratio=[1.3, 1.4],
random_area=[0.8, 0.9],
seed=1,
)
# coin1 = fn.random.coin_flip(dtype=types.DALIDataType.BOOL, seed=10)
# coin1_n = coin1 ^ True
# coin2 = fn.random.coin_flip(dtype=types.DALIDataType.BOOL, seed=20)
# coin2_n = coin2 ^ True
# images = (
# fn.flip(images, horizontal=1, vertical=1) * coin1 * coin2
# + fn.flip(images, horizontal=1) * coin1 * coin2_n
# + fn.flip(images, vertical=1) * coin1_n * coin2
# + images * coin1_n * coin2_n
# )
# flo = (
# fn.flip(flo, horizontal=1, vertical=1) * coin1 * coin2
# + fn.flip(flo, horizontal=1) * coin1 * coin2_n
# + fn.flip(flo, vertical=1) * coin1_n * coin2
# + flo * coin1_n * coin2_n
# )
# _flo = flo
# flo_0 = fn.slice(_flo, axis_names="C", start=0, shape=1)
# flo_1 = fn.slice(_flo, axis_names="C", start=1, shape=1)
# flo_0 = flo_0 * coin1 * -1 + flo_0 * coin1_n
# flo_1 = flo_1 * coin2 * -1 + flo_1 * coin2_n
# # flo = noflip + vertical flip + horizontal flip + both_flip
# # A horizontal flip is around the vertical axis (switch left and right)
# # So for a vertical flip coin1 is activated and needs to give +1, coin2 is activated needs to give -1
# # for a horizontal flip coin1 is activated and needs to be -1, coin2_n needs +1
# # no flip coin coin1_n +1, coin2_n +1
# flo = fn.cat(flo_0, flo_1, axis_name="C")
""" NORMALIZE """
images = fn.crop_mirror_normalize(
images,
mean=[0, 0, 0, 0, 0, 0],
std=[255, 255, 255, 255, 255, 255])
images = fn.crop_mirror_normalize(
images,
mean=[0.45, 0.432, 0.411, 0.45, 0.432, 0.411],
std=[1, 1, 1, 1, 1, 1],
)
flo = fn.crop_mirror_normalize(
flo, mean=[0, 0], std=[args.div_flow, args.div_flow])
pipeline.set_outputs(images, flo)
return pipeline
def load_tfrecord(directory, batch_size, training):
tfrecord = []
tfrecord_idx = []
for f in os.listdir(directory):
fullpath = os.path.join(directory, f)
if not os.path.isfile(fullpath):
continue
if f.endswith(".tfrecord"):
tfrecord.append(fullpath)
if f.endswith(".idx"):
tfrecord_idx.append(fullpath)
tfrecord.sort()
tfrecord_idx.sort()
pipe = Pipeline(batch_size=batch_size, num_threads=32, device_id=0)
with pipe:
inputs = fn.tfrecord_reader(
path=tfrecord,
index_path=tfrecord_idx,
features={
"frame_one":
tfrec.FixedLenFeature((), tfrec.string, ""),
"frame_two":
tfrec.FixedLenFeature((), tfrec.string, ""),
"frame_three":
tfrec.FixedLenFeature((), tfrec.string, ""),
"frame_four":
tfrec.FixedLenFeature((), tfrec.string, ""),
"plus_one_position":
tfrec.FixedLenFeature([3], tfrec.float32, 0.0),
"plus_one_orientation":
tfrec.FixedLenFeature([3], tfrec.float32, 0.0),
"plus_two_position":
tfrec.FixedLenFeature([3], tfrec.float32, 0.0),
"plus_two_orientation":
tfrec.FixedLenFeature([3], tfrec.float32, 0.0),
"plus_three_position":
tfrec.FixedLenFeature([3], tfrec.float32, 0.0),
"plus_three_orientation":
tfrec.FixedLenFeature([3], tfrec.float32, 0.0),
"speed":
tfrec.FixedLenFeature([], tfrec.float32, 0.0),
})
frame1 = inputs["frame_one"]
frame1 = fn.image_decoder(frame1,
device="mixed",
output_type=types.RGB)
# frame1 = fn.resize(frame1, device="gpu", resize_shorter=256.)
frame1 = fn.crop_mirror_normalize(frame1,
device="gpu",
dtype=types.FLOAT,
mean=[0., 0., 0.],
std=[1., 1., 1.])
frame1 = fn.transpose(frame1, device="gpu", perm=[1, 2, 0])
frame2 = inputs["frame_two"]
frame2 = fn.image_decoder(frame2,
device="mixed",
output_type=types.RGB)
# frame2 = fn.resize(frame2, device="gpu", resize_shorter=256.)
frame2 = fn.crop_mirror_normalize(frame2,
device="gpu",
dtype=types.FLOAT,
mean=[0., 0., 0.],
std=[1., 1., 1.])
frame2 = fn.transpose(frame2, device="gpu", perm=[1, 2, 0])
position = inputs["plus_one_position"].gpu()
orientation = inputs["plus_one_orientation"].gpu()
speed = inputs["speed"].gpu()
image = fn.cat(frame1, frame2, device="gpu", axis=2)
pose = fn.cat(position, orientation, device="gpu", axis=0)
pipe.set_outputs(image, pose, speed)
# Define shapes and types of the outputs
shapes = ((batch_size, 480, 640), (batch_size, 6), (batch_size))
dtypes = (tf.float32, tf.float32)
# Create dataset
return dali_tf.DALIDataset(pipeline=pipe,
batch_size=batch_size,
output_shapes=shapes,
output_dtypes=dtypes,
device_id=0)
