search

JinluZhang1126 / MixSTE

Official implementation of CVPR 2022 paper(MixSTE: Seq2seq Mixed Spatio-Temporal Encoder for 3D Human Pose Estimation in Video)
192 stars 16 forks source link

The visualization result is too small #33

Closed 2455DD closed 6 months ago

2455DD commented 1 year ago

After pulling the newest commit(620c9aaef474bf21677d40807b395282ef7f5fb2) and trying visualization, I find that the reconstruction result is too small compared with GT. output

My command is python run.py -k cpn_ft_h36m_dbb -c checkpoint --evaluate best_epoch.bin --render --viz-subject S11 --viz-action Wal king --viz-camera 0 --viz-output output.gif --viz-size 3 --viz-downsample 2 --v iz-limit 60

And the log is

Evaluate! python run.py -k cpn_ft_h36m_dbb -c checkpoint --evaluate best_epoch.bin --render --viz-subject S11 --viz-action Walking --viz-camera 0 --viz-output output.gif --viz-size 3 --viz-downsample 2 --viz-limit 60 CUDA Device Count: 0 Namespace(actions='*', alpha=0.01, batch_size=1024, beta=2, bone_length_term=True, by_subject=False, checkpoint='checkpoint', checkpoint_frequency=20, compare=False, coverlr=False, cs=512, data_augmentation=True, dataset='h36m', dense=False, dep=8, depth=4, disable_optimizations=False, downsample=1, dropout=0.0, epochs=120, evaluate='best_epoch.bin', export_training_curves=False, ft=False, ftchk='epoch_330.pth', ftpath='checkpoint/exp13_ft2d', ftpostrf=False, gpu='0', keypoints='cpn_ft_h36m_dbb', learning_rate=4e-05, linear_channel_size=1024, linear_projection=False, log='log/default', lr_decay=0.99, lr_decay_gap=10000, min_loss=100000, no_eval=False, no_proj=False, nolog=False, number_of_frames=243, postrf=False, render=True, resume='', stride=1, subjects_test='S9,S11', subjects_train='S1,S5,S6,S7,S8', subjects_unlabeled='', subset=1, test_time_augmentation=True, viz_action='Walking', viz_bitrate=3000, viz_camera=0, viz_downsample=2, viz_export=None, viz_limit=60, viz_no_ground_truth=False, viz_output='output.gif', viz_size=3, viz_skip=0, viz_subject='S11', viz_video=None, warmup=1) Loading dataset... Preparing data... Loading 2D detections... INFO: Receptive field: 243 frames INFO: Trainable parameter count: 33.783811 Million Loading checkpoint checkpoint/best_epoch.bin This model was trained for 157 epochs INFO: Testing on 225676 frames Rendering... Loading evaluate checkpoint checkpoint/best_epoch.bin This model was trained for 157 epochs

xobeiotozi commented 1 year ago

Have you solved this problem?

2455DD commented 1 year ago

Have you solved this problem?

well, I didn't. Instead, I tried implementing a visualization code by myself using detectron2 and matplotlib.

xobeiotozi commented 1 year ago

Have you solved this problem?

well, I didn't. Instead, I tried implementing a visualization code by myself using detectron2 and matplotlib.

Have you trained on the humaneva dataset using this model

2455DD commented 1 year ago

Have you trained on the humaneva dataset using this model

Sorry, I only trained (and tested) it on human36m

jianlai123-123 commented 8 months ago

Hello, may I contact you? I have some questions about visualization that I'd like to consult with you. Looking forward to your response@2455DD

jianlai123-123 commented 8 months ago

你好，请问可视化所需的数据集怎么获取的，除了官网获取，还有别的渠道嘛，官网注册时间太长了，期待您的回复，谢谢。 @2455DD

jianlai123-123 commented 8 months ago

Can you show the final structure of your project's data directory? Thank you. Looking forward to your response, and thanks again @2455DD

JinluZhang1126 commented 8 months ago

你好，我也是通过官方申请的。目前对于image/video数据貌似没有替代渠道，我也没有分发数据的权限，抱歉。

2024年1月23日 18:11，jianlai123-123 @.***> 写道：

你好，请问可视化所需的数据集怎么获取的，除了官网获取，还有别的渠道嘛，官网注册时间太长了，期待您的回复，谢谢。 @2455DD https://github.com/2455DD — Reply to this email directly, view it on GitHub https://github.com/JinluZhang1126/MixSTE/issues/33#issuecomment-1905717517, or unsubscribe https://github.com/notifications/unsubscribe-auth/AJBGHYACSXV4QFQHRZ2S32TYP6EGJAVCNFSM6AAAAAAT2NYUF6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTSMBVG4YTONJRG4. You are receiving this because you are subscribed to this thread.

2455DD commented 8 months ago

你好，请问可视化所需的数据集怎么获取的，除了官网获取，还有别的渠道嘛，官网注册时间太长了，期待您的回复，谢谢。

@jianlai123-123 不好意思有段时间没看Github了，数据集获取是从实验室资产中获取的，不方便分发，非常抱歉

Can you show the final structure of your project's data directory? Thank you. Looking forward to your response, and thanks again @2455DD

关于这个问题，我也已经记不得当时的情况了，您要是感兴趣的话我已经将我原有的MixSTE fork公开化，看看能不能帮到你

jianlai123-123 commented 8 months ago

你好，我也是通过官方申请的。目前对于image/video数据貌似没有替代渠道，我也没有分发数据的权限，抱歉。 … 2024年1月23日 18:11，jianlai123-123 @.***> 写道： 你好，请问可视化所需的数据集怎么获取的，除了官网获取，还有别的渠道嘛，官网注册时间太长了，期待您的回复，谢谢。 @2455DD https://github.com/2455DD — Reply to this email directly, view it on GitHub <#33 (comment)>, or unsubscribe https://github.com/notifications/unsubscribe-auth/AJBGHYACSXV4QFQHRZ2S32TYP6EGJAVCNFSM6AAAAAAT2NYUF6VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTSMBVG4YTONJRG4. You are receiving this because you are subscribed to this thread.

好的，谢谢

jianlai123-123 commented 8 months ago

你好，请问可视化所需的数据集怎么获取的，除了官网获取，还有别的渠道嘛，官网注册时间太长了，期待您的回复，谢谢。

@jianlai123-123 不好意思有段时间没看Github了，数据集获取是从实验室资产中获取的，不方便分发，非常抱歉

Can you show the final structure of your project's data directory? Thank you. Looking forward to your response, and thanks again @2455DD

关于这个问题，我也已经记不得当时的情况了，您要是感兴趣的话我已经将我原有的MixSTE fork公开化，看看能不能帮到你

谢谢您

jianlai123-123 commented 8 months ago

你好，进行可视化所用的数据就是video数据是吗，然后运行类似您给出的命令就可以得到可视化结果了吗?期待您的回复，谢谢。 @2455DD

2455DD commented 7 months ago

你好，进行可视化所用的数据就是video数据是吗，然后运行类似您给出的命令就可以得到可视化结果了吗?期待您的回复，谢谢。 @2455DD

应该是的，记不太得了

jianlai123-123 commented 6 months ago

您好，请问您在非数据集的视频可视化过吗？如果可视化过，请问怎么可视化的。谢谢 @2455DD @JinluZhang1126

2455DD commented 6 months ago

您好，请问您在非数据集的视频可视化过吗？如果可视化过，请问怎么可视化的。谢谢 @2455DD @JinluZhang1126

有过，具体来说我先把3D骨骼点结果跑出来以后用我fork里skeleton_visualize.py的相关功能做的可视化，你可以参考一下