wassname/pyrobolearn
1
0
Fork 0
mirror of https://github.com/wassname/pyrobolearn.git synced 2026-09-09 11:31:38 +08:00
Code Issues Packages Projects Releases Wiki Activity

add dataset parsers

Browse Source
This commit is contained in:
Brian Delhaisse
2019-06-21 20:17:09 +02:00
parent 2598e973ab
commit 477c258cd7
20 changed files with 14147 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
pyrobolearn/datasets/README.md
View File
pyrobolearn/datasets/parsers/3d/README.md
+54
View File
@@ -0,0 +1,54 @@
## 3D datasets
This repo contains a brief description of 3D datasets that I could find online. It also contains the Python code to load, parse, and plot some of these datasets.
### List of datasets
Here is a non-exhaustive list of datasets that could be used:
* [**TUM Kitchen**: "The TUM Kitchen Data Set of Everyday Manipulation Activities for Motion Tracking and Action Recognition", 2009](https://ias.in.tum.de/dokuwiki/software/kitchen-activity-data)
* scenarios: "everyday manipulation activities in a natural kitchen environment"
* labeled segmented data: 51 DOF articulated human model, 3D joint positions, videos and labels
* [**CMU-MMAC**: "Guide to the Carnegie Mellon University Multimodal Activity database", 2009](http://kitchen.cs.cmu.edu/)
* scenarios: human activity of subjects performing the tasks involved in cooking and food preparation
* [**HumanEva**: "HumanEva: Synchronized Video and Motion Capture Dataset and Baseline Algorithm for Evaluation of Articulated Human Motion", 2010](http://humaneva.is.tue.mpg.de/)
* 6 scenarios: walking, jogging, gesturing, throwing and catching a ball, boxing, combo (=walking+jogging+balancing)
* nb of subjects: 4
* data: 2D and 3D pose estimation
* [**MPII Cooking**: "A database for fine grained activity detection of cooking activities", 2012](https://www.mpi-inf.mpg.de/departments/computer-vision-and-multimodal-computing/research/human-activity-recognition/mpii-cooking-activities-dataset/)
<img src="figures/MPIICooking-comparison-datasets.png" alt="Datasets" style="width: 400px;"/>
* nb of subjects: 12
* 65 fine-grained cooking activities: background activity, change temperature, cut apart, cut dice, cut in, cut off ends, cut out inside, cut slices, cut stripes, dry, fill water from tap, grate, lid: put on, lid: remove, mix, move from X to Y, open egg, open tin, open/close cupboard, open/close drawer, open/close fridge, open/close oven, package X, peel, plug in/out, pour, pull out, puree, put in bowl, put in pan/pot, put on bread/dough, put on cutting-board, put on plate, read, remove from package, rip open, scratch off, screw close, screw open, shake, smell, spice, spread, squeeze, stamp, stir, strew, take & put in cupboard, take & put in drawer, take & put in fridge, take & put in oven, take & put in spice holder, take ingredient apart, take out from cupboard, take out from drawer, take out from fridge, take out from oven, take out from spice holder, taste, throw in garbage, unroll dough, wash hands, wash objects, whisk, wipe clean
* labeled and segmented data: 2D body joint positions
* [**J-HMDB**: "Joint-annotated Human Motion Data Base", 2013](http://jhmdb.is.tue.mpg.de/)
<img src="figures/JHMDB-comparison-datasets.png" alt="Datasets" style="width: 400px;"/>
* 21 scenarios: brush hair, catch, clap, climb stairs, golf, jump, kick ball, pick, pour, pull-up, push, run, shoot
ball, shoot bow, shoot gun, sit, stand, swing baseball, throw, walk, wave
* labeled and segmented data: 36-55 clips per action class with each clip containing 15-40 frames + 2D joint positions. 10 body parts connected by 13 joints (shoulder, elbow, wrist, hip, knee, ankle, neck) and two landmarks (face and belly).
* [**Human 3.6M**: "Human3.6M: Large Scale Datasets and Predictive Methods for 3D Human Sensing in Natural Environments", 2014](http://vision.imar.ro/human3.6m/description.php)
* 3.6 million 3D human poses and corresponding images
* nb of subjects: 11
* 15 scenarios: direction, discussion, eating, activitie while seated, greeting, taking photo, posing, making purchases, smoking, waiting, walking, sitting on chair, talking on the phone, walking dog, walking together
* data: 3D joint positions and joint angles
* [**UTD-MHAD**: "UTD-MHAD: A Multimodal Dataset for Human Action Recognition utilizing a Depth Camera and a Wearable Inertial Sensor", 2015](http://www.utdallas.edu/~kehtar/UTD-MHAD.html)
* nb of subjects: 8
* 27 different actions: (1) right arm swipe to the left, (2) right arm swipe to the right, (3) right hand wave, (4) two hand front clap, (5) right arm throw, (6) cross arms in the chest, (7) basketball shoot, (8) right hand draw x, (9) right hand draw circle (clockwise), (10) right hand draw circle (counter clockwise), (11) draw triangle, (12) bowling (right hand), (13) front boxing, (14) baseball swing from right, (15) tennis right hand forehand swing, (16) arm curl (two arms), (17) tennis serve, (18) two hand push, (19) right hand knock on door, (20) right hand catch an object, (21) right hand pick up and throw, (22) jogging in place, (23) walking in place, (24) sit to stand, (25) stand to sit, (26) forward lunge (left foot forward), (27) squat (two arms stretch out).
* labeled and segmented data: 861 data sequences, 3D skeleton joint positions
* [**G3D**: "G3D: A Gaming Action Dataset and Real Time Action Recognition Evaluation Framework", 2012](http://dipersec.king.ac.uk/G3D/index.html)
* nb of subjects: 10
* 20 gaming actions: punch right, punch left, kick right, kick left, defend, golf swing, tennis swing forehand, tennis swing backhand, tennis serve, throw bowling ball, aim and fire gun, walk, run, jump, climb, crouch, steer a car, wave, flap and clap
* data: 3D joint positions
* [**CMU-Datasets**](mocap.cs.cmu.edu/search.php)
* dataset: very diverse, contains data such as sport activities (soccer, dancing, etc), everyday movements/activities, locomotion (walk/run, jump, etc), interaction with environment and other agents, human subjects imitating animal behaviors (such as snake, dog, chicken, etc), stretching, finegrained movements, and so on.
* data: 3D joint positions and videos.
* Others: MSRC-12 Kinect Gesture Datase, MPI HDM05 Motion Capture Database, MSR Action3D Database, VGG Human Pose Estimation datasets, etc.
### Python Code
The python code to load, parse, and plot (using animations) has been implemented for the following datasets:
* UTD-MHAD: see `utd-mhad.ipynb`
* CMU-Datasets: see `cmu.ipynb`
pyrobolearn/datasets/parsers/3d/cmu.ipynb
+1769
View File
File diff suppressed because one or more lines are too long
pyrobolearn/datasets/parsers/3d/data/CMU/02.asf
+338
View File
@@ -0,0 +1,338 @@
# AST/ASF file generated using VICON BodyLanguage
# -----------------------------------------------
:version 1.10
:name VICON
:units
mass 1.0
length 0.45
angle deg
:documentation
.ast/.asf automatically generated from VICON data using
VICON BodyBuilder and BodyLanguage model FoxedUp or BRILLIANT.MOD
:root
order TX TY TZ RX RY RZ
axis XYZ
position 0 0 0
orientation 0 0 0
:bonedata
begin
id 1
name lhipjoint
direction 0.655637 -0.713449 0.247245
length 2.52691
axis 0 0 0 XYZ
end
begin
id 2
name lfemur
direction 0.34202 -0.939693 0
length 7.59371
axis 0 0 20 XYZ
dof rx ry rz
limits (-160.0 20.0)
(-70.0 70.0)
(-60.0 70.0)
end
begin
id 3
name ltibia
direction 0.34202 -0.939693 0
length 7.28717
axis 0 0 20 XYZ
dof rx
limits (-10.0 170.0)
end
begin
id 4
name lfoot
direction 0.0886837 -0.243657 0.965798
length 2.2218
axis -90 7.62852e-016 20 XYZ
dof rx rz
limits (-45.0 90.0)
(-70.0 20.0)
end
begin
id 5
name ltoes
direction 1.53547e-011 -4.22004e-011 1
length 1.11249
axis -90 7.62852e-016 20 XYZ
dof rx
limits (-90.0 20.0)
end
begin
id 6
name rhipjoint
direction -0.645062 -0.722004 0.250209
length 2.49697
axis 0 0 0 XYZ
end
begin
id 7
name rfemur
direction -0.34202 -0.939693 0
length 7.58734
axis 0 0 -20 XYZ
dof rx ry rz
limits (-160.0 20.0)
(-70.0 70.0)
(-70.0 60.0)
end
begin
id 8
name rtibia
direction -0.34202 -0.939693 0
length 7.21538
axis 0 0 -20 XYZ
dof rx
limits (-10.0 170.0)
end
begin
id 9
name rfoot
direction -0.102907 -0.282735 0.953662
length 2.23735
axis -90 -7.62852e-016 -20 XYZ
dof rx rz
limits (-45.0 90.0)
(-20.0 70.0)
end
begin
id 10
name rtoes
direction -1.5354e-011 -4.21751e-011 1
length 1.11569
axis -90 -7.62852e-016 -20 XYZ
dof rx
limits (-90.0 20.0)
end
begin
id 11
name lowerback
direction 0.00952301 0.997604 -0.0685217
length 2.05943
axis 0 0 0 XYZ
dof rx ry rz
limits (-20.0 45.0)
(-30.0 30.0)
(-30.0 30.0)
end
begin
id 12
name upperback
direction 0.00494399 0.999577 -0.0286701
length 2.06523
axis 0 0 0 XYZ
dof rx ry rz
limits (-20.0 45.0)
(-30.0 30.0)
(-30.0 30.0)
end
begin
id 13
name thorax
direction 0.000729686 0.999992 0.00388603
length 2.06807
axis 0 0 0 XYZ
dof rx ry rz
limits (-20.0 45.0)
(-30.0 30.0)
(-30.0 30.0)
end
begin
id 14
name lowerneck
direction 0.00452882 0.995459 0.0950816
length 1.57426
axis 0 0 0 XYZ
dof rx ry rz
limits (-20.0 45.0)
(-30.0 30.0)
(-30.0 30.0)
end
begin
id 15
name upperneck
direction 0.021922 0.997711 -0.0639762
length 1.56399
axis 0 0 0 XYZ
dof rx ry rz
limits (-20.0 45.0)
(-30.0 30.0)
(-30.0 30.0)
end
begin
id 16
name head
direction 0.008024 0.999444 -0.0323691
length 1.6265
axis 0 0 0 XYZ
dof rx ry rz
limits (-20.0 45.0)
(-30.0 30.0)
(-30.0 30.0)
end
begin
id 17
name lclavicle
direction 0.967826 0.247107 -0.0474446
length 3.6598
axis 0 0 0 XYZ
dof ry rz
limits (-20.0 10.0)
(0.0 20.0)
end
begin
id 18
name lhumerus
direction 1 -4.48971e-011 -2.90544e-027
length 4.86513
axis 180 -30 -90 XYZ
dof rx ry rz
limits (-60.0 90.0)
(-90.0 90.0)
(-90.0 90.0)
end
begin
id 19
name lradius
direction 1 -4.48958e-011 1.79638e-026
length 3.35554
axis 180 -30 -90 XYZ
dof rx
limits (-10.0 170.0)
end
begin
id 20
name lwrist
direction 1 -4.48952e-011 3.34345e-026
length 1.67777
axis -2.22354e-014 90 90 XYZ
dof ry
limits (-180.0 0.0)
end
begin
id 21
name lhand
direction 1 -4.49055e-011 6.6869e-026
length 0.661175
axis -4.26681e-014 90 90 XYZ
dof rx rz
limits (-90.0 90.0)
(-45.0 45.0)
end
begin
id 22
name lfingers
direction 1 -4.48731e-011 1.33738e-025
length 0.533057
axis -8.53362e-014 90 90 XYZ
dof rx
limits (0.0 90.0)
end
begin
id 23
name lthumb
direction 0.707107 -6.35005e-011 0.707107
length 0.765366
axis -90 45 2.85299e-015 XYZ
dof rx rz
limits (-45.0 45.0)
(-45.0 45.0)
end
begin
id 24
name rclavicle
direction -0.973174 0.211421 -0.0907389
length 3.59444
axis 0 0 0 XYZ
dof ry rz
limits (-10.0 20.0)
(-20.0 0.0)
end
begin
id 25
name rhumerus
direction -1 -4.48964e-011 1.56311e-027
length 5.02649
axis 180 30 90 XYZ
dof rx ry rz
limits (-90.0 60.0)
(-90.0 90.0)
(-90.0 90.0)
end
begin
id 26
name rradius
direction -1 -4.48948e-011 2.01724e-026
length 3.36431
axis 180 30 90 XYZ
dof rx
limits (-10.0 170.0)
end
begin
id 27
name rwrist
direction -1 -4.49008e-011 3.34345e-026
length 1.68216
axis -2.22354e-014 -90 -90 XYZ
dof ry
limits (-180.0 0.0)
end
begin
id 28
name rhand
direction -1 -4.49013e-011 6.6869e-026
length 0.730406
axis -4.26681e-014 -90 -90 XYZ
dof rx rz
limits (-90.0 90.0)
(-45.0 45.0)
end
begin
id 29
name rfingers
direction -1 -4.48867e-011 1.33738e-025
length 0.588872
axis -8.53362e-014 -90 -90 XYZ
dof rx
limits (0.0 90.0)
end
begin
id 30
name rthumb
direction -0.707107 -6.34907e-011 0.707107
length 0.845506
axis -90 -45 -2.85299e-015 XYZ
dof rx rz
limits (-45.0 45.0)
(-45.0 45.0)
end
:hierarchy
begin
root lhipjoint rhipjoint lowerback
lhipjoint lfemur
lfemur ltibia
ltibia lfoot
lfoot ltoes
rhipjoint rfemur
rfemur rtibia
rtibia rfoot
rfoot rtoes
lowerback upperback
upperback thorax
thorax lowerneck lclavicle rclavicle
lowerneck upperneck
upperneck head
lclavicle lhumerus
lhumerus lradius
lradius lwrist
lwrist lhand lthumb
lhand lfingers
rclavicle rhumerus
rhumerus rradius
rradius rwrist
rwrist rhand rthumb
rhand rfingers
end
pyrobolearn/datasets/parsers/3d/data/CMU/02_01.amc
+10293
View File
File diff suppressed because it is too large Load Diff
pyrobolearn/datasets/parsers/3d/data/CMU/download-link.txt
+3
View File
@@ -0,0 +1,3 @@
# CMU Dataset
* webpage (link to download the dataset): mocap.cs.cmu.edu/search.php
pyrobolearn/datasets/parsers/3d/data/CMU/subject2.txt
+15
View File
@@ -0,0 +1,15 @@
Subject #2 (various expressions and human behaviors)
02_01 02_01.tvd 02_01.c3d 02_01.amc 02_01.avi walk
02_02 02_02.tvd 02_02.c3d 02_02.amc 02_02.mpg 02_02.avi walk
02_03 02_03.tvd 02_03.c3d 02_03.amc 02_03.mpg 02_03.avi run/jog
02_04 02_04.tvd 02_04.c3d 02_04.amc 02_04.mpg 02_04.avi jump, balance
02_05 02_05.tvd 02_05.c3d 02_05.amc 02_05.mpg 02_05.avi punch/strike
02_06 02_06.tvd 02_06.c3d 02_06.amc 02_06.mpg 02_06.avi bend over, scoop up, rise, lift arm
02_07 02_07.tvd 02_07.c3d 02_07.amc 02_07.mpg 02_07.avi swordplay
02_08 02_08.tvd 02_08.c3d 02_08.amc 02_08.mpg 02_08.avi swordplay
02_09 02_09.tvd 02_09.c3d 02_09.amc 02_09.mpg 02_09.avi swordplay
02_10 02_10.tvd 02_10.c3d 02_10.amc 02_10.mpg 02_10.avi wash self
pyrobolearn/datasets/parsers/3d/data/UTD/Skeleton_joint_order.txt
+24
View File
@@ -0,0 +1,24 @@
The skeleton joint order in UTD-MAD dataset:
1. head;
2. shoulder_center;
3. spine;
4. hip_center;
5. left_shoulder;
6. left_elbow;
7. left_wrist;
8. left_hand;
9. right_shoulder;
10. right_elbow;
11. right_wrist;
12. right_hand;
13. left_hip;
14. left_knee;
15. left_ankle;
16. left_foot;
17. right_hip;
18. right_knee;
19. right_ankle;
20. right_foot;
Each skeleton data is a 20 x 3 x num_frame matrix. Each row of a skeleton frame corresponds to three spatial coordinates of a joint.
pyrobolearn/datasets/parsers/3d/data/UTD/download-link.txt
+3
View File
@@ -0,0 +1,3 @@
# UTD-MHAD Dataset
* webpage (link to download the dataset): http://www.utdallas.edu/~kehtar/UTD-MHAD.html
pyrobolearn/datasets/parsers/3d/figures/JHMDB-comparison-datasets.png
BIN
View File
Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 90 KiB

pyrobolearn/datasets/parsers/3d/figures/MPIICooking-comparison-datasets.png
BIN
View File
Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 108 KiB

pyrobolearn/datasets/parsers/3d/mocap_parser.py
+230
View File
@@ -0,0 +1,230 @@
import numpy as np
from scipy.interpolate import CubicSpline
import matplotlib.pyplot as plt
class MocapParser(object):
def __init__(self, filename):
"""
Parser for motion capture. By default, if the data is described in Cartesian space, the x-axis should
be pointing in front of the human, the y-axis on his/her left, and z-axis upward.
:param filename:
"""
self.filename = filename
self.num_samples = 0
self.joint_names = []
self.link_names = []
self.marker_names = []
self.data = self.loadFile(filename)
def loadFile(self, filename):
raise NotImplementedError("loadFile is not implemented.")
def interpolate(self, data, method='cubic', axis=-1)
"""
Interpolate the Mocap data such that it is between 0 and 1, along the given axis.
:param data: mocap data
:param method: 'linear', 'cubic', 'hermite' interpolation
:param axis: The axis on which to interpolate. The length should be equal to the number of samples in the
mocap data
:return: Interpolator - function that given the time [0,1] will give the corresponding data
"""
self.num_samples = data.shape[axis]
x = np.linspace(0., 1., self.num_samples)
interpolator = CubicSpline(x, self.data, axis=axis)
return interpolator
def getMarkerName(self, marker_idx=None):
if marker_idx is None:
return self.getMarkerNames()
else:
return self.marker_names[marker_idx]
def getMarkerNames(self):
return self.marker_names
def getJointName(self, joint_idx=None):
if joint_idx is None:
return self.getJointNames()
else:
return self.joint_names[joint_idx]
def getJointNames(self):
return self.joint_names
def getLinkName(self, link_idx=None):
if link_idx is None:
return self.getLinkNames()
else:
return self.link_names[link_idx]
def getLinkNames(self):
return self.link_names
def getMarkerPosition(self, marker_idx=None):
if marker_idx is None:
return self.getMarkerPositions()
else:
pass
def getMarkerPositions(self):
pass
def getJointPosition(self, joint_idx=None):
if joint_idx is None:
return self.getJointPositions()
else:
pass
def getJointPositions(self):
pass
def getJointVelocity(self, joint_idx=None):
if joint_idx is None:
return self.getJointVelocities()
else:
pass
def getJointVelocities(self):
pass
def getLinkPosition(self, link_idx=None):
if link_idx is None:
return self.getLinkPositions()
else:
pass
def getLinkPositions(self):
pass
def getLinkVelocity(self, link_idx=None):
if link_idx is None:
return self.getLinkVelocities()
else:
pass
def getLinkVelocities(self):
pass
def getLinkOrientation(self, link_idx=None):
if link_idx is None:
return self.getLinkOrientations()
else:
pass
def getLinkOrientations(self):
pass
def getLinkAngularVelocity(self, link_idx=None):
if link_idx is None:
return self.getLinkAngularVelocities()
else:
pass
def getLinkAngularVelocities(self):
pass
## Plotting ##
def plot3d(self, ax=None):
pass
def plotJointProfile(self, ax=None, joint_idx=None, pos=True, vel=True, acc=True):
pass
def plotLinkProfile(self, ax=None, link_idx=None, pos=True, vel=True, acc=True, wrt='world'):
pass
def plotMarkerProfile(self, ax=None, link_idx=None, pos=True, vel=True, acc=True, wrt='world'):
pass
def animate3d(self, ax=None, title=None):
pass
from amcparser.skeleton import Skeleton
from amcparser.motion import SkelMotion
class CMUMocapParser(MocapParser):
def __init__(self, skeleton_filename, motion_filename, skeleton_scale=1.0):
super(CMUMocapParser, self).__init__(motion_filename)
self.joint_names = ['head', 'upperneck', 'lowerneck', 'upperback', 'thorax', 'lowerback', 'root', # Spine
'rclavicle', 'rhumerus', 'rradius', 'rwrist', 'rhand', 'rthumb', 'rfingers', # Right arm
'lclavicle', 'lhumerus', 'lradius', 'lwrist', 'lhand', 'lthumb', 'lfingers', # Left arm
'rhipjoint', 'rfemur', 'rtibia', 'rfoot', 'rtoes', # Right leg
'lhipjoint', 'lfemur', 'ltibia', 'lfoot', 'ltoes'] # Left leg
self.link_names = self.joint_names
self.marker_names = self.joint_names
self.base_name = 'root'
# Load skeleton
self.skeleton = Skeleton(skeleton_filename, scale=skeleton_scale)
def loadFile(self, filename, framerate=120.):
self.skeleton_motion = SkelMotion(self.skeleton, filename, (1./framerate))
# compute trajectories
#self.data = self.skeleton_motion.traverse(bone, start, end)
self.data = self.skeleton_motion.traverse(None, 0, -1)
# make sure that given axis
def animate3d(self, ax=None, title=None):
if ax is None:
fig = plt.figure()
ax = fig.gca(projection='3d')
# Rescaling such that the skeleton it is in the right proportion and at the middle
xmin, xmax = X[..., 2].min(), X[..., 2].max()
ymin, ymax = X[..., 0].min(), X[..., 0].max()
zmin, zmax = X[..., 1].min(), X[..., 1].max()
x_len, y_len, z_len = (xmax - xmin), (ymax - ymin), (zmax - zmin)
max_len = max([x_len, y_len, z_len])
xmin, xmax = xmin + (x_len - max_len) / 2., xmin + (x_len + max_len) / 2.
ymin, ymax = ymin + (y_len - max_len) / 2., ymin + (y_len + max_len) / 2.
zmin, zmax = zmin + (z_len - max_len) / 2., zmin + (z_len + max_len) / 2.
# Plot trajectories
x, y, z = skel.bones['rhand'].xyz_data.T
T = len(x)
def init():
ax.set_title('movement')
ax.set_xlabel('x')
ax.set_xlim(xmin, xmax)
ax.set_ylabel('y')
ax.set_ylim(ymin, ymax)
ax.set_zlabel('z')
ax.set_zlim(zmin, zmax)
# ax.scatter(x[0], y[0], z[0], marker='o')
return fig,
def animate(i):
# ax.view_init(elev=10., azim=i)
ax.scatter(x[i], y[i], z[i], marker='o')
return fig,
def animate_skeleton(i):
ax.clear()
init()
# ax.scatter(X[:,2,i], X[:,0,i], X[:,1,i], marker='o', c='b')
for d in [X_TO, X_RA, X_LA, X_RL, X_LL]:
ax.plot(d[:, i, 2], d[:, i, 0], d[:, i, 1], marker='o', c='b')
return [fig] # fig,
# Animate
# anim = animation.FuncAnimation(fig, animate, init_func=init,
# frames=T, interval=20, blit=True)
anim = animation.FuncAnimation(fig, animate_skeleton, init_func=init,
frames=T, interval=20, blit=False)
plt.show()
# Test
if __name__ == "__main__":
pass
pyrobolearn/datasets/parsers/3d/utd-mhad.ipynb
+967
View File
File diff suppressed because one or more lines are too long
pyrobolearn/datasets/parsers/handwritten/HWRT/download-link.txt
+4
View File
@@ -0,0 +1,4 @@
# HWRT Database of Handwritten symbols
* webpage: www.martin-thoma.de/write-math/data/
* link to download the dataset: i13pc106.ira.uka.de/~mthoma/hwrt/2015-01-28-data.tar
pyrobolearn/datasets/parsers/handwritten/IAM-Online/download-link.txt
+7
View File
@@ -0,0 +1,7 @@
# IAM Online Handwriting Database
* webpage: www.fki.inf.unibe.ch/databases/iam-on-line-handwriting-database
* link to download the dataset: http://www.fki.inf.unibe.ch/databases/iam-on-line-handwriting-database/download-the-iam-on-line-handwriting-database
Note that you need to register in order to download the datasets.
pyrobolearn/datasets/parsers/handwritten/ICDAR2013/download-link.txt
+5
View File
@@ -0,0 +1,5 @@
# ICDAR2013 - Handwriting Stroke Recovery from Offline Data (Kaggle)
* webpage: https://www.kaggle.com/c/icdar2013-stroke-recovery-from-offline-data
* link to download the dataset: https://www.kaggle.com/c/icdar2013-stroke-recovery-from-offline-data/data
pyrobolearn/datasets/parsers/handwritten/README.md
+38
View File
@@ -0,0 +1,38 @@
## Online Handwritten Datasets
This repo contains the python code to load, parse and plot online handwritten characters.
In this folder, I focus on the following datasets:
1. Character Trajectories Data Set (2,858): https://archive.ics.uci.edu/ml/datasets/Character+Trajectories
2. UJI Pen Characters (V2) Data Set (11,640): archive.ics.uci.edu/ml/datasets/UJI+Pen+Characters+(Version+2)
3. HWRT Database of Handwritten symbols: www.martin-thoma.de/write-math/data/
4. ICDAR2013 - Handwriting Stroke Recovery from Offline Data (Kaggle): https://www.kaggle.com/c/icdar2013-stroke-recovery-from-offline-data
5. IAM Online Handwriting Database: www.fki.inf.unibe.ch/databases/iam-on-line-handwriting-database
There are of course other datasets (mostly offline) that could be used such as:
* MNIST dataset (Offline)
* NIST Handprinted Forms and Characters Database (Offline)
* The IRESTE On/Off Dual Handwriting Database (75€): www.irccyn.ec-nantes.fr/~viardgau/IRONOFF/ICDAR99.htm
* UCI online datasets:
* Online Handwritten Assamese Characters Dataset (8,235): archive.ics.uci.edu/ml/datasets/Online+Handwritten+Assamese+Characters+Dataset
* Pen-Based Recognition of Handwritten Digits Dataset (10,992): archive.ics.uci.edu/ml/datasets/Pen-Based+Recognition+of+Handwritten+Digits
* CASIA online and offline chinese handwriting Databases: www.nlpr.ia.ac.cn/databases/handwriting/Online_database.html
* CEDAR Handwriting (Offline)
## How to use it?
The whole python code to load and parse the different datasets is inside the jupyter notebook `handwritten-data.ipynb`. To open the jupyter notebook, you will first need to install jupyter. For each dataset, I plotted 25 random samples to give you a general feeling about the dataset. You will need to download the dataset that you which to use (and put it in the corresponding folder if you do not wish to modify the python code).
## FAQs
* What is the meaning of 'online'? Online in this case means that you also have the trajectories (x,y), while offline would mean that you only have pictures of handwritten characters/words/sentences.
* Do I need to cite the dataset that I am using? Yes, you should cite the corresponding paper which can usually be found on the website where you downloaded the dataset.
* For what can I use these datasets? Mostly for demonstration / supervised learning. You could, for instance, use one of the following models to learn the trajectories: Gaussian Processes, Gaussian Mixture Models, Hidden Markov Models, (Deep) Neural Networks, Dynamic Movement Primitives, etc.
pyrobolearn/datasets/parsers/handwritten/UCI-Characters-Trajectories/download-link.txt
+5
View File
@@ -0,0 +1,5 @@
# Character Trajectories Data Set
* webpage: https://archive.ics.uci.edu/ml/datasets/Character+Trajectories
* link to download the dataset: https://archive.ics.uci.edu/ml/machine-learning-databases/character-trajectories/
pyrobolearn/datasets/parsers/handwritten/UCI-Ujipen-Chars2/download-link.txt
+4
View File
@@ -0,0 +1,4 @@
# UJI Pen Characters (V2) Data Set
* webpage: archive.ics.uci.edu/ml/datasets/UJI+Pen+Characters+(Version+2)
* link to download the dataset: http://archive.ics.uci.edu/ml/machine-learning-databases/uji-penchars/version2/
pyrobolearn/datasets/parsers/handwritten/handwritten-data.ipynb
+388
View File
File diff suppressed because one or more lines are too long