stanfordnmbl / opencap-processing Goto Github PK

Failing with kids and not returning correct error:

No peaks are detected but code going through, crashing later.

To reproduce:
session_id = '16467055-c264-40e7-8a56-a3945ea2d729' # 1.25m/s
trial_name = 'kids19'

https://simtk.org/plugins/phpBB/viewtopicPhpbb.php?f=2385&t=16096&p=0&start=0&view=&sid=d20ec46cb68326e438a2723204f50e74

The problem is that the provided end time for the time window exceeds the end time of the motion file.

Support parallel computing.

getpass doesn't work in spyder. Plus, figure out how to send 2fa, and potentially store credentials for the duration of the session, but reliably delete them once spyder is closed

It seems like the libraries build on OS 12.5 do not work with older versions:https://simtk.org/plugins/phpBB/viewtopicPhpbb.php?f=2385&t=15341&p=44584&start=0&view=

Add get_whole_body_angular_momentum to kinematics class.

Bug when coordinate files are provided in rad.

They return modelName with and without .osim

Seems like it is not picking the good compiler by default when installing Visual Studio 2022. Check if there is a way to identify the compiler on the computer and specify it in the cmake calls.

Manual fix:

• Change this line to cmd1 = 'cmake "' + pathBuildExpressionGraph + '" -G "Visual Studio 17 2022" -A x64 -DTARGET_NAME:STRING="' + filename + '" -DSDK_DIR:PATH="' + SDK_DIR + '" -DCPP_DIR:PATH="' + CPP_DIR + '"'
• Change this line to cmd3 = 'cmake "' + pathBuildExternalFunction + '" -G "Visual Studio 17 2022" -A x64 -DTARGET_NAME:STRING="' + filename + '" -DINSTALL_DIR:PATH="' + path_external_functions_filename_install + '"'

Installing opensim, condacolab the sourcecode from github runs without problems on colab. I ran it directly and aslo made copy to an colab notebook version which gives me the same error.

I get following errors when importing packages
ModuleNotFoundError Traceback (most recent call last)
in
1 import os
----> 2 import utilsKinematics
3 from utils import download_kinematics
4 from utilsPlotting import plot_dataframe

/content/opencap-processing/utilsKinematics.py in
20
21 import os
---> 22 import opensim
23 import numpy as np
24 import pandas as pd

ModuleNotFoundError: No module named 'opensim'

Bug when running kinetics with no .env file.

I suggest we add a function get_gait_events in the gait_analysis class that returns gaitEvents. That would be useful for downstream tasks.

Thoughts @suhlrich? I can do it.

• Segmentation: the algo assumes the end of a squat is the beginning of the next one. This isn't good if there are long pauses between reps, should identify beginning and end independently.
• Single leg squat: identify leg of interest
  • Separate left/right leg metrics
  • Only calculate ipsilateral leg for single leg
• Trunk lean wrt ground
• Trunk lean wrt pelvis
• KA if available
• ankle eversion (check Matt DeMers study)
• integrate into web app

Some squats for testing

https://simtk.org/plugins/phpBB/viewtopicPhpbb.php?f=2385&t=16705&p=0&start=0&view=&sid=928c54de5ed0ee4c00b88db99a78a496

1. HRNet is too confident, and when the person is half out of the field of view, it still gives good confidence to keypoints but the results are garbage, need to take that into account.
2. For results to be meaningful over time, we should alway pick the same leg. Let's make this an optional flag that we can then use when deploying.

Check wrapping surface scaling thing.

Hard to get good fitting for this session.
ef516897-f0b2-493e-9927-3022db2d2ac3

Improve documentation - test conda package https://anaconda.org/anaconda/cmake

A lot of the functions in utils will download / request even if it is already available locally.

For instance, download_trial will call get_model_and_metadata that will call get_neutral_trial_id that will call get_session_json. This makes the system not usable offline and can create bugs if the API is temporarily unavailable (happened to me overnight). It would be great if we had a consistent way of defining these functions. For each function would take an argument overwrite=False, such that by default it does not download again, but instead uses what is there already.

Thoughts @suhlrich ?

I get following issue while running the download data script no module named 'decouple'. Any tips how to resolved this? Has it something to do with the installating of !pip install python-decouple? There were no errors while installing packages.

• clean up code
• example treadmill
• example gait
• example STS
• example squats
• example DJ
• outputs
• plotting
• documentation
• testing on Scott machine
• testing on mac
• testing on linux
• authentication spyder
• authentication pycharm
• authentication terminal
• option to write an env file
• visual studio 2017, 2019, 2022 (looks like there is no need to specify compiler, to confirm but that's great)
• deprecation warning
• joint reaction analysis
• treadmill vs non-treadmill is same session
• bounds hip adds

@suhlrich feel free to add items

The issue links to this discussion in the forum.

I would like to know if it will be possible to include a full body model with complete spine musculature (ex. FBLS) to work with opencap processing?

As several other users have pointed out in the forum, adding support for frontal knee ROM in kinematic and kinetic analyses would be a great feature to implement soon.

I am referring to the model posted here: https://drive.google.com/file/d/153QJQHVuYFVj01p_NKJTlD_Ho570N84L/view

Best

Bernhard

Not a big fan that different filters are used for inputs and outputs. The OpenSim filter is a 3rd order I think

Add class utilsKinetics for analyzing kinetics results

Hello, I keep getting the attribution error like the one below.

AttributeError: module 'foo' has no attribute 'foo'

It is coming from the generateF function in utilsOpenSimAD.py

def generateF(dim):
import foo
importlib.reload(foo)
cg = ca.CodeGenerator('foo_jac')
arg = ca.SX.sym('arg', dim)
y,, = foo.foo(arg)
F = ca.Function('F',[arg],[y])
cg.add(F)
cg.add(F.jacobian())
cg.generate()

How can I solve this problem?

I appreciate your answer in advance.

I followed the tutorial and did the set-up as instructed. The example.py ran successfully. Immediately, I tried running example_kineitcs.py, these two errors occurred:

Add otp verification step in utilsAuth in opencap-core and opencap-processing. If otp_verified is False, we can trigger the code to be sent just like we do for the front end, and have the user enter it into the terminal, and move on.

Add support for running muscle-driven simulations with the full body model with the ISB shoulder.

Full body model sent by user.

Bug here, you cannot assign a NaN to an array of doubles. Need for fix + need for way to handle missing contralateral data in rest of code. Session for testing: 1a1e9c15-4bc3-4b72-bf9d-cc6ba9293c23, trial: 10mwt

Automatically identify cmake-generator on windows

Test out new feature for bounds

optionssol.detect_simple_bounds = true; % (default = false)
opti.solver('ipopt',optionssol);
sol = opti.solve(); (edited)

It is all hard coded now, let's define the bounds only once and adjust everything automatically.

1. Step length is calculating a norm from heel to heel instead of y position.
2. gait frame is non orthogonal. replace with

        z_temp = np.zeros((self.nGaitCycles,3))
        for i in range(self.nGaitCycles):
            z_temp[i,:] = np.mean(asisVector[self.gaitEvents['ipsilateralIdx'][i,0]: \
                             self.gaitEvents['ipsilateralIdx'][i,2]],axis=0)
        z_temp = z_temp / np.linalg.norm(z_temp,axis=1,keepdims=True)
        ```

• Segmentation. Algo is clearly not robust (eg, assumes two peaks will be detected). While accuracy sure matters, I don't think any metric will be computed at a certain instant in time but rather max over contact period for instance. So it should not matter too much if we are slightly off. I used marker data for now, it seems a little off when comparing to the video/simulation so open to anything else. Also wasn't sure for now whether to use right vs left data for segmentation, or an average or the first in contact, or ... Open to suggestion.
• Support single-leg
  • Segmentation
  • Identification
• Add peak KA if available
• Do we report both right and left separately for double leg dropjumps? An average? Peaks?
• Trunk lean, I saw that you computed that differently, might be worth checking which method is best