General Information¶
This notebook demotstraits transformation
https://goodok.github.io/fastai_sparse/overview/augmentation.svg
Imports¶
In [1]:
import numpy as np
from pathlib import Path
from matplotlib import pyplot as plt
from PIL import Image as pImage
from ipywidgets import VBox, HBox, Layout
from ipywidgets import widgets
import io
from fastai_sparse import utils, visualize
from fastai_sparse.utils import log
from fastai_sparse.data_items import MeshItem, PointsItem, SparseItem
from fastai_sparse import transforms as T
# autoreload python modules on the fly when its source is changed
%load_ext autoreload
%autoreload 2
Experiment environment¶
In [2]:
experiment_name = 'transforms'
In [3]:
utils.watermark(pandas=True)
virtualenv: (fastai_sparse) python: 3.6.8 nvidia driver: b'384.130' nvidia cuda: 9.0, V9.0.176 cudnn: 7.1.4 torch: 1.0.1.post2 pandas: 0.24.2 fastai: 1.0.48 fastai_sparse: 0.0.4.dev0
Notebook options¶
In [4]:
utils.wide_notebook()
# uncomment this lines if you want switch off interactive and save visaulisation as screenshoots:
# For rendering run command in terminal: `chromium-browser --remote-debugging-port=9222`
IS_INTERACTIVE = False
if not IS_INTERACTIVE:
visualize.options.interactive = False
visualize.options.save_images = True
visualize.options.verbose = True
visualize.options.filename_pattern_image = Path('images', experiment_name, 'fig_{fig_number}')
Utils¶
In [5]:
def transform_and_show(x, transforms,
original_point_size=1.4, result_point_size=1.4):
b = x.copy().apply_tfms(transforms)
b.describe()
img_x = x.show(width=480, height=600, point_size_value=original_point_size)
img_b = b.show(width=480, height=600, point_size_value=result_point_size)
if IS_INTERACTIVE:
r = HBox([VBox([widgets.HTML('<div style="text-align: center; width: 100%; font-size:16px">Original</div>'), img_x]),
VBox([widgets.HTML('<div style="text-align: center; width: 100%; font-size:16px">Transformed</div>'), img_b])
])
return r
else:
fig, axes = plt.subplots(ncols=2, figsize=(21, 15))
ax = axes[0]
img = pImage.open(io.BytesIO(img_x.data))
ax.imshow(img)
ax.set_title('Original', fontsize=14)
ax.axis('off')
ax = axes[1]
img = pImage.open(io.BytesIO(img_b.data))
ax.imshow(img)
ax.set_title('Transformed', fontsize=14)
ax.axis('off')
plt.subplots_adjust(wspace=0.0, hspace=0)
plt.show()
Read example¶
read mesh represantation from file
In [6]:
SOURCE_DIR = Path('..', 'samples', 'scannet')
assert SOURCE_DIR.exists()
In [7]:
fn = SOURCE_DIR / 'scene0000_00' / 'scene0000_00.merged.ply'
assert fn.exists()
In [8]:
o = MeshItem.from_file(fn)
o.describe()
MeshItem (scene0000_00.merged.ply) vertices: shape: (81369, 3) dtype: float64 min: -0.01657, max: 8.74040, mean: 3.19051 faces: shape: (153587, 3) dtype: int64 min: 0, max: 81368, mean: 40549.68796 colors: shape: (81369, 4) dtype: uint8 min: 1.00000, max: 255.00000, mean: 145.80430 labels: shape: (81369,) dtype: uint16 min: 0.00000, max: 230.00000, mean: 12.97057 Colors from vertices Labels from vertices
In [9]:
o.show(with_normals=True, point_size_value=0.4)
Saved to file: images/transforms/fig_1.png
Out[9]:
Combined transformation¶
mesh 🡢 points cloud 🡢 sparse voxels
In [10]:
tfms = [
T.rotate(),
T.rotate_XY(),
T.flip_x(p=0.5),
T.rand_translate(offset=(-2, 2, 3)),
T.rand_scale(scale=[0.7, 1.3]),
T.fit_to_sphere(),
T.translate(offset=1),
T.scale(scale=50),
T.to_points_cloud(method='vertices'),
T.sample_points(num_points=100000),
T.elastic(gran=8, mag=64),
T.merge_features(ones=False, normals=True, colors=True),
T.to_sparse_voxels(),
]
In [11]:
np.random.seed(46)
In [12]:
transform_and_show(o, tfms)
id: scene0000_00.merged coords shape: (81369, 3) dtype: int64 min: 7, max: 100, mean: 53.43846 features shape: (81369, 7) dtype: float32 min: -0.99999, max: 255.00000, mean: 83.26527 x shape: (81369,) dtype: int64 min: 14, max: 100, mean: 59.20743 y shape: (81369,) dtype: int64 min: 7, max: 92, mean: 52.17837 z shape: (81369,) dtype: int64 min: 34, max: 81, mean: 48.92958 labels shape: (81369,) dtype: int64 min: 0, max: 230, mean: 12.96269 voxels: 13183 points / voxels: 6.172267313964955
Saved to file: images/transforms/fig_2.png
Saved to file: images/transforms/fig_3.png
Mesh transforms¶
affine¶
In [13]:
np.random.seed(44)
In [14]:
affine = [
T.rotate(),
T.rotate_XY(),
T.flip_x(p=0.5),
T.rand_translate(offset=(-2, 2, 3)),
T.rand_scale(scale=[0.7, 1.3]),
T.fit_to_sphere(),
T.translate(offset=1),
T.scale(scale=50),
]
transform_and_show(o, affine)
MeshItem (scene0000_00.merged.ply) vertices: shape: (81369, 3) dtype: float64 min: 7.70902, max: 93.38925, mean: 50.00000 faces: shape: (153587, 3) dtype: int64 min: 0, max: 81368, mean: 40549.68796 colors: shape: (81369, 4) dtype: uint8 min: 1.00000, max: 255.00000, mean: 145.80430 labels: shape: (81369,) dtype: uint16 min: 0.00000, max: 230.00000, mean: 12.97057 Colors from vertices Labels from vertices
Saved to file: images/transforms/fig_4.png
Saved to file: images/transforms/fig_5.png
Points cloud transforms¶
In [15]:
np.random.seed(44)
In [16]:
pc = o.copy().apply_tfms(affine)
pc = pc.apply_tfms([T.to_points_cloud(method='vertices')])
#pc.show()
elastic¶
In [17]:
np.random.seed(40)
In [18]:
transform_and_show(pc, [T.elastic(gran=8, mag=48)])
PointsItem (scene0000_00.merged) points shape: (81369, 3) dtype: float64 min: 5.52502, max: 94.71804, mean: 49.73212 labels shape: (81369,) dtype: uint16 min: 0.00000, max: 230.00000, mean: 12.97057 colors shape: (81369, 4) dtype: uint8 min: 1.00000, max: 255.00000, mean: 145.80430 normals shape: (81369, 3) dtype: float32 min: -1.00000, max: 0.99997, mean: -0.01491
Saved to file: images/transforms/fig_6.png
Saved to file: images/transforms/fig_7.png
Color transform¶
noise¶
In [19]:
np.random.seed(0)
In [20]:
transform_and_show(pc, [T.colors_normalize(), T.colors_noise()])
PointsItem (scene0000_00.merged) points shape: (81369, 3) dtype: float32 min: 7.70902, max: 93.38925, mean: 50.00000 labels shape: (81369,) dtype: uint16 min: 0.00000, max: 230.00000, mean: 12.97057 colors shape: (81369, 3) dtype: float32 min: -0.92077, max: 1.17641, mean: -0.03715 normals shape: (81369, 3) dtype: float32 min: -1.00000, max: 0.99997, mean: -0.01491
Saved to file: images/transforms/fig_8.png
Saved to file: images/transforms/fig_9.png
Convertation¶
from mesh to points cloud¶
In [21]:
transform_and_show(o, [T.to_points_cloud(method='vertices')],
original_point_size=0.2, result_point_size=0.2)
PointsItem (scene0000_00.merged) points shape: (81369, 3) dtype: float32 min: -0.01657, max: 8.74040, mean: 3.19051 labels shape: (81369,) dtype: uint16 min: 0.00000, max: 230.00000, mean: 12.97057 colors shape: (81369, 4) dtype: uint8 min: 1.00000, max: 255.00000, mean: 145.80430 normals shape: (81369, 3) dtype: float32 min: -1.00000, max: 1.00000, mean: 0.06483
Saved to file: images/transforms/fig_10.png
Saved to file: images/transforms/fig_11.png
from points cloud to sparse voxels represantation¶
In [22]:
transform_and_show(pc, [T.merge_features(colors=True), T.to_sparse_voxels()])
id: scene0000_00.merged coords shape: (81369, 3) dtype: int64 min: 7, max: 93, mean: 49.49963 features shape: (81369, 5) dtype: float32 min: 1.00000, max: 255.00000, mean: 116.84344 x shape: (81369,) dtype: int64 min: 7, max: 93, mean: 49.49961 y shape: (81369,) dtype: int64 min: 11, max: 80, mean: 49.50094 z shape: (81369,) dtype: int64 min: 16, max: 93, mean: 49.49832 labels shape: (81369,) dtype: int64 min: 0, max: 230, mean: 12.97057 voxels: 18269 points / voxels: 4.453938365537249
Saved to file: images/transforms/fig_12.png
Saved to file: images/transforms/fig_13.png
