search

FORTH-ModelBasedTracker / MocapNET

We present MocapNET, a real-time method that estimates the 3D human pose directly in the popular Bio Vision Hierarchy (BVH) format, given estimations of the 2D body joints originating from monocular color images. Our contributions include: (a) A novel and compact 2D pose NSRM representation. (b) A human body orientation classifier and an ensemble of orientation-tuned neural networks that regress the 3D human pose by also allowing for the decomposition of the body to an upper and lower kinematic hierarchy. This permits the recovery of the human pose even in the case of significant occlusions. (c) An efficient Inverse Kinematics solver that refines the neural-network-based solution providing 3D human pose estimations that are consistent with the limb sizes of a target person (if known). All the above yield a 33% accuracy improvement on the Human 3.6 Million (H3.6M) dataset compared to the baseline method (MocapNET) while maintaining real-time performance
https://www.youtube.com/watch?v=Jgz1MRq-I-k
Other
845 stars 135 forks source link

can’t work ./MocapNET2LiveWebcamDemo There was a problem with the file initialize.sh #121

Open notbugnotwork opened 3 months ago

notbugnotwork commented 3 months ago

sudo ./MocapNET2LiveWebcamDemo --from /mnt/c/Users/Administrator/Downloads/test.mp4 --openpose 2024-06-10 12:00:00.529398: W tensorflow/stream_executor/platform/default/dso_loader.cc:59] Could not load dynamic library 'libcudart.so.10.1'; dlerror: libcudart.so.10.1: cannot open shared object file: No such file or directory 2024-06-10 12:00:00.529467: I tensorflow/stream_executor/cuda/cudart_stub.cc:29] Ignore above cudart dlerror if you do not have a GPU set up on your machine. CPU : Intel(R) Xeon(R) CPU E3-1231 v3 @ 3.40GHz GPU : Trying to open source (/mnt/c/Users/Administrator/Downloads/test.mp4) Using BVH codebase version 0.74 buildOpenGLProjectionForIntrinsics: img(720x1280) fx 582.18, fy 582.53, cx 360.00, cy 640.00 Fast BVH parser revealed 212 joints / 0 motion fields

Number of Values Per Frame: 498 Allocating 1992 bytes of memory to hold BVH motions Frames: 1(1) / Frame Time : 0.0400 BVH subsystem initialized using dataset/headerWithHeadAndOneMotion.bvh that contains 1 frames ( 498 motions each ) Initializing IK code.. Body Problem :

The IK problem "Body" we want to solve is ultimately divided into 8 kinematic chains Chain 0 has 20 parts : jID(hip/0)->mID(0 to 2) jID(hip/0)->mID(3 to 5) jID(neck1/4)->EndEffector jID(head/5)->EndEffector jID(EndSite_eye.l/71)->EndEffector jID(EndSite_eye.r/79)->EndEffector jID(temporalis02.l/94)->EndEffector jID(temporalis02.r/99)->EndEffector jID(rShldr/105)->EndEffector jID(rForeArm/106)->EndEffector jID(rHand/107)->EndEffector jID(lShldr/136)->EndEffector jID(lForeArm/137)->EndEffector jID(lHand/138)->EndEffector jID(rThigh/167)->EndEffector jID(rShin/168)->EndEffector jID(rFoot/169)->EndEffector jID(lThigh/190)->EndEffector jID(lShin/191)->EndEffector jID(lFoot/192)->EndEffector Chain 1 has 6 parts : jID(neck1/4)->mID(15 to 17) jID(head/5)->EndEffector jID(EndSite_eye.l/71)->EndEffector jID(EndSite_eye.r/79)->EndEffector jID(temporalis02.l/94)->EndEffector jID(temporalis02.r/99)->EndEffector Chain 2 has 3 parts : jID(rShldr/105)->mID(237 to 239) jID(rForeArm/106)->mID(240 to 242) jID(rHand/107)->EndEffector Chain 3 has 3 parts : jID(lShldr/136)->mID(315 to 317) jID(lForeArm/137)->mID(318 to 320) jID(lHand/138)->EndEffector Chain 4 has 5 parts : jID(rThigh/167)->mID(393 to 395) jID(rShin/168)->mID(396 to 398) jID(rFoot/169)->mID(399 to 401) jID(EndSite_toe1-2.R/172)->EndEffector jID(EndSite_toe5-3.R/188)->EndEffector Chain 5 has 3 parts : jID(rFoot/169)->mID(399 to 401) jID(EndSite_toe1-2.R/172)->EndEffector jID(EndSite_toe5-3.R/188)->EndEffector Chain 6 has 5 parts : jID(lThigh/190)->mID(447 to 449) jID(lShin/191)->mID(450 to 452) jID(lFoot/192)->mID(453 to 455) jID(EndSite_toe1-2.L/195)->EndEffector jID(EndSite_toe5-3.L/211)->EndEffector Chain 7 has 3 parts : jID(lFoot/192)->mID(453 to 455) jID(EndSite_toe1-2.L/195)->EndEffector jID(EndSite_toe5-3.L/211)->EndEffector End of problem "Body" BVH code initalization successfull.. Trying to set thread realtime prio = 99 SCHED_FIFO OK Thread priority is now 99 Graph dataset/combinedModel/openpose_model.pb could not be found

i don't konw how to fix it ,it seems like the MocapNET2LiveWebcamDemo is binary file,i can't figure out

lf-cardoso commented 3 months ago

I faced the same problem and found that for some reason, initialize.sh couldn't download and extract the weights.

I resolved it by manually downloading and extracting the allInOneMNET2RedistMirrorICPR2020.zip (drive.google.com/u/3/uc?id=1GtmPWOpf3MzhqhqegaC8cS3_m3Drp6y3&export=download&confirm=yes) and CMUPlusHeadMotionCapture.zip (drive.google.com/u/3/uc?id=1Zt-MycqhMylfBUqgmW9sLBclNNxoNGqV&export=download&confirm=yes) files. Then, I added the MotionCapture directory from the CMUPlusHeadMotionCapture.zip file to the dataset directory.

notbugnotwork commented 3 months ago

When I use wget --no-check-certificate to download a file manually, it works without any issues. However, when I replace wget with wget --no-check-certificate in the script and rerun it, the script still fails to download the files. Almost all the parts of the script that involve wget are unable to download the files.

notbugnotwork commented 3 months ago

我遇到了同样的问题，发现由于某种原因，initialize.sh 无法下载和提取权重。

我通过手动下载并提取 allInOneMNET2RedistMirrorICPR2020.zip (drive.google.com/u/3/uc?id=1GtmPWOpf3MzhqhqegaC8cS3_m3Drp6y3&export=download&confirm=yes) 和 CMUPlusHeadMotionCapture.zip (drive.google.com/u/3/uc?id=1Zt-MycqhMylfBUqgmW9sLBclNNxoNGqV&export=download&confirm=yes) 文件解决了这个问题。然后，我将 CMUPlusHeadMotionCapture.zip 文件中的 MotionCapture 目录添加到数据集目录中。

This issue might require someone knowledgeable about scripts to resolve.

AmmarkoV commented 3 months ago

This is a server problem, probably related to https://github.com/FORTH-ModelBasedTracker/MocapNET/issues/106

I will contact the network administrators of FORTH to try and solve this..