open-mmlab / mmpose Goto Github PK

tried to run demo

image size in config file is single integer but in _box2cs() expecting sequantial

I use higher_hrnet32_coco_512x512.py config

ref open-mmlab/mmaction2#209

I am getting very slow performance on a single person video on RTX2080Ti with the following command

python demo/bottom_up_video_demo.py configs/bottom_up/resnet/coco/res152_coco_512x512.py res152_coco_512x512-364eb38d_20200822.pth --video-path demo/tabata1_640x360.mp4 --show --device cuda:0

FPS ~2.5. It was calculated as total processing time divided by the number of video frames, so it is not super precise...

Any idea what might be wrong?

Looking at the viz_pose_results funciton used in topdown image demo it looks like this is specifically for 17 keypoint pose estimation. Is there another function that needs to be used to visualize hand poses or is this not implemented yet?

Are the model support for COCO-WholeBody dataset?

I followed get_start.md , download model form https://mmpose.readthedocs.io/en/latest/model_zoo.html model_name is pose_resnet_50 and img_size is 156*192
use
python -m torch.distributed.launch --nproc_per_node=8 tools/test.py configs/top_down/resnet/coco/res50_coco_256x192.py checkpoints/res50_coco_256x192-ec54d7f3_20200709.pth --launcher pytorch --eval mAP

but got
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets= 20 ] = 0.735
Average Precision (AP) @[ IoU=0.50 | area= all | maxDets= 20 ] = 0.925
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets= 20 ] = 0.814
Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets= 20 ] = 0.706
Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets= 20 ] = 0.779
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 20 ] = 0.765
Average Recall (AR) @[ IoU=0.50 | area= all | maxDets= 20 ] = 0.934
Average Recall (AR) @[ IoU=0.75 | area= all | maxDets= 20 ] = 0.833
Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets= 20 ] = 0.733
Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets= 20 ] = 0.814
so whats wrong?

在推断完成后，该如何拿到处理后的结果数据，而不仅仅是一张图片？

Hello guys.
Thank you for your great contribution.

I tried some top-down method (model zoo by using the scripts inside mmpose/demo, and it works well.

Unfortunately, I can't run the bottom-up model yet.
Do you guys have any clue how to do this?

Thank you.

Hi, buddies, In my virtual env, I run a test demo: ./tools/dist_test.sh configs/top_down/resnet/coco/res50_coco_256x192.py checkpoints/res50_coco_256x192-ec54d7f3_20200709.pth 2 --eval mAP
and I got a env problems: ImportError: cannot import name 'EpochBasedRunner' from 'mmcv.runner' (/home/stelo/anaconda3/envs/mmpose/lib/python3.7/site-packages/mmcv/runner/init.py)
should I rebuild my env? Or conda remove mmcv packages and reinstall it?

A very nice work!

I have some questions about the benchmark.

1. In the benchmark table, I found that with the same input size, all the models of res50, 101, 152, hrnet-w32, and hrnet-w48 have almost the same speed. However, all these models have different GFLOPs. Could you explain why this happens?
2. Comparing with the implementation of HRNet, the training speed of hrnet serious backbone has a big improvement. Could you provide more details about where the improvement comes from?

Another suggestion about the benchmark. Maybe you can consider also provide the inference speed.

Ref: https://github.com/open-mmlab/mmdetection/blob/master/.github/workflows/build.yml

Hello,

Thank you very much for training the whole body COCO model.

Could you please check if the link to dark_pose_hrnet_w48+ is correct?

Here is the download link I found in https://mmpose.readthedocs.io/en/latest/top_down_models.html

The config name is hrnet_w48_coco_wholebody_384x288_dark, but the model name that I got is hrnet_w32_coco_256x192-c78dce93_20200708.pth

Thanks

ref

https://github.com/open-mmlab/mmaction2/pull/25/files

https://github.com/open-mmlab/mmaction2/pull/204/files

Hi，how about the pose benchmark of using shufflenet_v2 as backbone?

Hi, I can not achieve the reported training speed for multiple configs.

The followings are my test.

4 NVidia V100SXM2 (16G memory)
2 x Intel(R) Xeon(R) Gold 6148 CPU @ 2.40GHz

用自底向上的方法，是否需要手工对待测试的视频裁成正方形，才能得到更好的性能？

How to show pose label like this.

And If it can, Can I train pose? not object.
For example, MMdetction will show three person labels in the above image.
But I want to show another three labels in the above image like sitting, hitting and standing.

Version 0.6 will be released soon. Now it's time to plan for the next iteration, i.e. v0.7. This will be a one-month iteration and our focus in this iteration is to enrich the supported tasks. We will add support for 3D shapes, and add more hand tasks.

Algorithms:

• HMR (3D human shape recovery) (#162)

Datasets:

• LSP (not needed any more)
• H36M (#159)
• COCO-WholeBody (#78)
• Frei-hand (#125)
• Panotic (#144)

Tools:

• Hand demo (#115)

Misc:

I simply tested the speed.
GPU: T4
There are two people in the test video.

mmpose:
res50_coco_256x192: 27 FPS
mobilenetv2_coco_256x192: 31 FPS

alphapose:
fast_res50_256x192: 50 FPS

Why is there such a big gap in keypoint_head.

Firstly, thanks for sharing the mmpose as open source project.

The code is successfully installed on my PC and the demo runs successfully. However, when I try to run scripts/test.py in order to check the mAP on mscoco dataset, the error happened with FileNotFoundError: [Errno 2] No such file or directory: 'data/coco/person_detection_results/COCO_val2017_detections_AP_H_56_person.json'. I thought when I the .json file should be generated when the demo run. Am I right ? Is there any suggestion that how can I find or generate this .json file ?

Thank you and Best wishes

OSError: hrnet_w48_coco_256x192/epoch_210.pth is not a checkpoint file

it has cost almost 1 hours ...and it seems that it doesn't train...
does any body meet this case before? many thanks..........

log_level = 'INFO'
load_from = None
resume_from = None
dist_params = dict(backend='nccl')
workflow = [('train', 1)]
checkpoint_config = dict(interval=10)
evaluation = dict(interval=10, metric='mAP')

optimizer = dict(
    type='Adam',
    lr=5e-4,
)
optimizer_config = dict(grad_clip=None)
# learning policy
lr_config = dict(
    policy='step',
    warmup='linear',
    warmup_iters=500,
    warmup_ratio=0.001,
    step=[170, 200])
total_epochs = 210
log_config = dict(
    interval=50,
    hooks=[
        dict(type='TextLoggerHook'),
        # dict(type='TensorboardLoggerHook')
    ])

channel_cfg = dict(
    num_output_channels=17,
    dataset_joints=17,
    dataset_channel=[
        [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16],
    ],
    inference_channel=[
        0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16
    ])

# model settings
model = dict(
    type='TopDown',
    pretrained='models/pytorch/imagenet/'
    'mobilenet_v2_batch256_20200708-3b2dc3af.pth',
    backbone=dict(type='MobileNetV2', widen_factor=1., out_indices=(7, )),
    keypoint_head=dict(
        type='TopDownSimpleHead',
        in_channels=1280,
        out_channels=channel_cfg['num_output_channels'],
    ),
    train_cfg=dict(),
    test_cfg=dict(
        flip_test=True,
        post_process=True,
        shift_heatmap=True,
        unbiased_decoding=False,
        modulate_kernel=11),
    loss_pose=dict(type='JointsMSELoss', use_target_weight=True))

data_cfg = dict(
    image_size=[192, 256],
    heatmap_size=[48, 64],
    num_output_channels=channel_cfg['num_output_channels'],
    num_joints=channel_cfg['dataset_joints'],
    dataset_channel=channel_cfg['dataset_channel'],
    inference_channel=channel_cfg['inference_channel'],
    soft_nms=False,
    nms_thr=1.0,
    oks_thr=0.9,
    vis_thr=0.2,
    bbox_thr=1.0,
    use_gt_bbox=False,
    image_thr=0.0,
    bbox_file='data/coco/person_detection_results/'
    'COCO_val2017_detections_AP_H_56_person.json',
)

train_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(type='TopDownRandomFlip', flip_prob=0.5),
    dict(
        type='TopDownHalfBodyTransform',
        num_joints_half_body=8,
        prob_half_body=0.3),
    dict(
        type='TopDownGetRandomScaleRotation', rot_factor=40, scale_factor=0.5),
    dict(type='TopDownAffine'),
    dict(type='ToTensor'),
    dict(
        type='NormalizeTensor',
        mean=[0.485, 0.456, 0.406],
        std=[0.229, 0.224, 0.225]),
    dict(type='TopDownGenerateTarget', sigma=2),
    dict(
        type='Collect',
        keys=['img', 'target', 'target_weight'],
        meta_keys=[
            'image_file', 'joints_3d', 'joints_3d_visible', 'center', 'scale',
            'rotation', 'bbox_score', 'flip_pairs'
        ]),
]

valid_pipeline = [
    dict(type='LoadImageFromFile'),
    dict(type='TopDownAffine'),
    dict(type='ToTensor'),
    dict(
        type='NormalizeTensor',
        mean=[0.485, 0.456, 0.406],
        std=[0.229, 0.224, 0.225]),
    dict(
        type='Collect',
        keys=[
            'img',
        ],
        meta_keys=[
            'image_file', 'center', 'scale', 'rotation', 'bbox_score',
            'flip_pairs'
        ]),
]

test_pipeline = valid_pipeline

data_root = '/share/jwh/utils/data/zipped/object/detect/coco/coco2017'
data = dict(
    samples_per_gpu=64,
    workers_per_gpu=2,
    train=dict(
        type='TopDownCocoDataset',
        ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
        img_prefix=f'{data_root}/val2017/',
        data_cfg=data_cfg,
        pipeline=train_pipeline),
    val=dict(
        type='TopDownCocoDataset',
        ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
        img_prefix=f'{data_root}/val2017/',
        data_cfg=data_cfg,
        pipeline=valid_pipeline),
    test=dict(
        type='TopDownCocoDataset',
        ann_file=f'{data_root}/annotations/person_keypoints_val2017.json',
        img_prefix=f'{data_root}/val2017/',
        data_cfg=data_cfg,
        pipeline=valid_pipeline),
)

For now, we only restrict the minimum version of mmcv.

Restricting upper bound sounds attractive but it might invites trouble when you want to use multiple repos in mm-series.

Thanks for the extraordinary work!

As I am trying to test on my own data, I have found that if the bounding box scores in coco format json is saved as 1 (instead of 1.0), there will be the info => Total boxes after filter low [email protected]: 0 and test does not start. I changed the scores to 0.999 and it works.

Is this deliberate, or is it inherited from pycocotools?

Hello,
I am trying to find a way to do batch inference with mmpose, but it seems that the project has not supported it yet. There is only a function called "_inference_single_pose_model" inside inference.py.

Is there any quick way I can adapt mmpose to batch inference? I think it's an essential feature because most of the time, we just work on bach inference. real-time infernece with a single input image is not possible with the current architectures, so single-image inference is kind of not useful.

Thank you and I look forward to hearing from you soon.
Best.

excuse me, there's some error when i try to use demo/img_demo.py
would you please give me some advices, thank you very very much.
truly sorry for wasting your time.

1. use hrnet_w32_coco_384x288.py
   + hrnet_w32_coco_384x288-d9f0d786_20200708.pth
   on coco2017val

(mmpose) :/work/wjw_2020/app/mmpose-master_20200716/mmpose$ python demo/img_demo.py
configs/top_down/hrnet/coco/hrnet_w32_coco_384x288.py
work/hrnet_w32_coco_384x288-d9f0d786_20200708.pth
--img-root data/coco/val2017/
--json-file data/coco/annotations/person_keypoints_val2017.json
--show

loading annotations into memory...
Done (t=0.24s)
creating index...
index created!
Traceback (most recent call last):
File "demo/img_demo.py", line 91, in
main()
File "demo/img_demo.py", line 72, in main
pose_model, image_name, person_bboxes, format='xywh')
File "/home/gy/work/wjw_2020/app/mmpose-master_20200716/mmpose/mmpose/apis/inference.py", line 230, in inference_pose_model
pose = _inference_single_pose_model(model, img_or_path, bbox)
File "/home/gy/work/wjw_2020/app/mmpose-master_20200716/mmpose/mmpose/apis/inference.py", line 151, in _inference_single_pose_model
test_pipeline = [LoadImage()] + cfg.test_pipeline[1:]
File "/home/gy/work/wjw_2020/app/mmpose-master_20200716/mmpose/mmcv/mmcv/utils/config.py", line 349, in getattr
return getattr(self._cfg_dict, name)
File "/home/gy/work/wjw_2020/app/mmpose-master_20200716/mmpose/mmcv/mmcv/utils/config.py", line 37, in getattr
raise ex
AttributeError: 'ConfigDict' object has no attribute 'test_pipeline'

2. use hrnet_w32_coco_384x288.py
   + hrnet_w32_coco_384x288-d9f0d786_20200708.pth
   on test folder

(mmpose):/work/wjw_2020/app/mmpose-master_20200716/mmpose$ python demo/img_demo.py \
configs/top_down/hrnet/coco/hrnet_w32_coco_384x288.py
work/hrnet_w32_coco_384x288-d9f0d786_20200708.pth
--img-root tests/data/
--json-file tests/data/test_coco.json
--show

loading annotations into memory...
Done (t=0.00s)
creating index...
index created!
Traceback (most recent call last):
File "demo/img_demo.py", line 91, in
main()
File "demo/img_demo.py", line 72, in main
pose_model, image_name, person_bboxes, format='xywh')
File "/home/gy/work/wjw_2020/app/mmpose-master_20200716/mmpose/mmpose/apis/inference.py", line 230, in inference_pose_model
pose = _inference_single_pose_model(model, img_or_path, bbox)
File "/home/gy/work/wjw_2020/app/mmpose-master_20200716/mmpose/mmpose/apis/inference.py", line 151, in _inference_single_pose_model
test_pipeline = [LoadImage()] + cfg.test_pipeline[1:]
File "/home/gy/work/wjw_2020/app/mmpose-master_20200716/mmpose/mmcv/mmcv/utils/config.py", line 349, in getattr
return getattr(self._cfg_dict, name)
File "/home/gy/work/wjw_2020/app/mmpose-master_20200716/mmpose/mmcv/mmcv/utils/config.py", line 37, in getattr
raise ex
AttributeError: 'ConfigDict' object has no attribute 'test_pipeline'

3-1.configs/bottom_up/higherhrnet/coco/higher_hrnet32_coco_512x512.py
+ higher_hrnet32_coco_512x512-8ae85183_20200713.pth
same error: AttributeError: 'ConfigDict' object has no attribute 'test_pipeline'

3-2.configs/top_down/resnet/coco
+ res50_coco_256x192-ec54d7f3_20200709.pth
same error: AttributeError: 'ConfigDict' object has no attribute 'test_pipeline'

4.other infomations:
4-1.ubuntu16.04, 1080Ti, Python 3.7.7, pytorch 1.5.1, GCC 7.3.0, mmcv 1.0.2, numpy 1.16.4, opencv 4.3.0, json_tricks 3.15.2
4-2.CUDA Version: 10.1 Driver Version: 418.87.00
4-3.get mmpose at 2020.07.16 by: git clone https://github.com/open-mmlab/mmpose.git
4-4.get mmcv at 2020.07.16 by: pip install mmcv
same error after try to uninstall and 'pip install mmcv-full';
same error after try to uninstall and 'pip install mmcv-full==latest+torch1.5.0+cu101 -f https://openmmlab.oss-accelerate.aliyuncs.com/mmcv/dist/index.html' command in https://github.com/open-mmlab/mmcv,
same error after try to uninstall and 'git clone https://github.com/open-mmlab/mmcv.git && cd mmcv && pip install -e .'
same error after try to uninstall and 'git clone https://github.com/open-mmlab/mmcv.git && cd mmcv && MMCV_WITH_OPS=1 pip install -e .'

4-6.conda list (it's too long, sorry!):
Name Version Build Channel
_libgcc_mutex 0.1 main
addict 2.2.1
albumentations 0.4.6
appdirs 1.4.4
blas 1.0 mkl
ca-certificates 2020.6.24 0
certifi 2020.6.20 py37_0
cityscapesScripts 1.5.0
clang-format 9.0.0
cudatoolkit 10.1.243 h6bb024c_0
cycler 0.10.0
Cython 0.29.21
decorator 4.4.2
freetype 2.10.2 h5ab3b9f_0
future 0.18.2
imagecorruptions 1.1.0
imageio 2.9.0
imgaug 0.4.0
intel-openmp 2020.1 217
jpeg 9b 0
json-tricks 3.15.2
kiwisolver 1.2.0
ld_impl_linux-64 2.33.1 h53a641e_7
libedit 3.1.20191231 h7b6447c_0
libffi 3.3 he6710b0_2
libgcc-ng 9.1.0 hdf63c60_0
libgfortran-ng 7.3.0 hdf63c60_0
libpng 1.6.37 hbc83047_0
libstdcxx-ng 9.1.0 hdf63c60_0
libtiff 4.1.0 h2733197_1
lz4-c 1.9.2 he6710b0_0
matplotlib 3.2.2
mkl 2020.1 217
mkl_fft 1.0.12 py37ha843d7b_0
mkl_random 1.1.1 py37h0573a6f_0
mmcv 1.0.2
mmcv-full 1.0.2
mmcv-nightly 1.0rc0.dev20200630
mmdet 2.3.0rc0+5f2cded
mmpose 0.1.0+unknown
munkres 1.0.7 py_1
ncurses 6.2 he6710b0_1
networkx 2.4
ninja 1.7.2 0
numpy 1.16.4 py37h7e9f1db_0
numpy-base 1.16.4 py37hde5b4d6_0
olefile 0.46 py_0
opencv-python 4.3.0.36
openssl 1.1.1g h7b6447c_0
Pillow 6.2.2
pillow 7.1.2 py37hb39fc2d_0
pip 20.1.1 py37_1
pycocotools 2.0
pyparsing 2.4.7
python 3.7.7 hcff3b4d_5
python-dateutil 2.8.1
pytorch 1.5.1 py3.7_cuda10.1.243_cudnn7.6.3_0 pytorch
PyWavelets 1.1.1
PyYAML 5.3.1
readline 8.0 h7b6447c_0
scikit-image 0.17.2
scipy 1.5.1
setuptools 47.3.1 py37_0
Shapely 1.7.0
six 1.15.0
sqlite 3.32.3 h62c20be_0
terminaltables 3.1.0
tifffile 2020.7.4
tk 8.6.10 hbc83047_0
torchvision 0.6.1 py37_cu101 pytorch
wheel 0.34.2 py37_0
xz 5.2.5 h7b6447c_0
yapf 0.30.0
zlib 1.2.11 0
zstd 1.4.4 h0b5b093_3

4-7.one of the config i use,P.S.: have not do any change:
hrnet_w32_coco_384x288.py.txt

4-8.one of the weights i use,
hrnet_w32_coco_384x288-d9f0d786_20200708.pth from https://mmpose.readthedocs.io/en/latest/model_zoo.html

md5: 105a8076e55599d9e7ab0026eb894326

4-9.other infomation:
this enviroment running mmdet normally/ok;
i have search 'AttributeError: 'ConfigDict' object has no attribute 'test_pipeline'' with bing; then found 2 issue in mmdetection's issues, it said that the issue-asker's config is too old and git lastest mmdet can solve his problem.

maybe i did some wrong operation on enviroment...
look forward to reply,thankyou very much

Hello, thank you very much for creating a very helpful pose estimation project.
I am trying to fine-tune the model configs/top_down/hrnet/coco/hrnet_w48_coco_384x288.py with the pre-trained weights: hrnet_w48_coco_384x288-314c8528_20200708.pth on the posetrack dataset, and the below warning show up the console.

The training command line I used is:
python tools/train.py ./configs/top_down/hrnet_mvai/coco/hrnet_w48_coco_384x288.py

And here is the pre-trained weight that I used.

And here is the warning message:

unexpected key in source state_dict: backbone.conv1.weight, backbone.bn1.weight, backbone.bn1.bias, backbone.bn1.running_mean, backbone.bn1.running_var, backbone.bn1.num_batches_tracked, backbone.conv2.weight, backbone.bn2.weight, backbone.bn2.bias, backbone.bn2.running_mean, backbone.bn2.running_var, backbone.bn2.num_batches_tracked, backbone.layer1.0.conv1.weight, backbone.layer1.0.bn1.weight, etc

missing keys in source state_dict: conv1.weight, bn1.weight, bn1.bias, bn1.running_mean, bn1.running_var, conv2.weight, bn2.weight, bn2.bias, bn2.running_mean, bn2.running_var, layer1.0.conv1.weight, layer1.0.bn1.weight, layer1.0.bn1.bias, layer1.0.bn1.running_mean, layer1.0.bn1.running_var, layer1.0.conv2.weight, layer1.0.bn2.weight, layer1.0.bn2.bias, layer1.0.bn2.running_mean, layer1.0.bn2.running_var, layer1.0.conv3.weight, layer1.0.bn3.weight, layer1.0.bn3.bias, layer1.0.bn3.running_mean, layer1.0.bn3.running_var, layer1.0.downsample.0.weight, layer1.0.downsample.1.weight, layer1.0.downsample.1.bias, layer1.0.downsample.1.running_mean, layer1.0.downsample.1.running_var, layer1.1.conv1.weight, layer1.1.bn1.weight, layer1.1.bn1.bias, layer1.1.bn1.running_mean, layer1.1.bn1.running_var, layer1.1.conv2.weight, layer1.1.bn2.weight, layer1.1.bn2.bias, layer1.1.bn2.running_mean, layer1.1.bn2.running_var, layer1.1.conv3.weight, layer1.1.bn3.weight, etc

Hi everybody

Currently only one sample per GPU is possible while executing tests, which leads to very slow validation and testing for me. I assume to fix this the forward_test methods for the TopDown and BottomUp classes would need to be updated. Is this planned anytime soon or should I give it a try myself to contribute?

Ｈow to test on an image and get the whole body result (an image with whole body pose estimation)?

Version 0.6 will be released soon. Now it's time to plan for the next iteration, i.e. v0.7. This will be a one-month iteration.

Algorithms:

Datasets:

Tools:

hi,
thx for your great work,

Im using configs/top_down/resnet/coco/res50_coco_256x192.py to train my own keypoint dataset(git mmpose at 2020.08.06), theres some questions i met:

[line72--73] have been changed to my own file:

bbox_file='data/person_detection_results/'
'bbox_1000pic.json',

[line126--137] changed to:

    train=dict(
    type='TopDownCocoDataset',
    ann_file=f'{data_root}/annotations/person_keypoints_train_1000pic.json',
    img_prefix=f'{data_root}/train_1000pic/',
    data_cfg=data_cfg,
    pipeline=train_pipeline),

    val=dict(
    type='TopDownCocoDataset',
    ann_file=f'{data_root}/annotations/person_keypoints_train_1000pic.json',
    img_prefix=f'{data_root}/train_1000pic/',
    data_cfg=data_cfg,
    pipeline=valid_pipeline),

In this setting, everything is alright, train on 1000pic successfully, valid on 1000pic successfully, output result.json in work_dir successfully;

but when i change valid, for example, when valid part changed from:

    ann_file=f'{data_root}/annotations/person_keypoints_train_1000pic.json',
    img_prefix=f'{data_root}/train_1000pic/',

to

    ann_file=f'{data_root}/annotations/person_keypoints_valid_20pic.json'
    img_prefix=f'{data_root}/valid_20pic/',

it seems mmpose read bbox_1000pic.json from line72--73, and consider there should be 1000images to valid, then read valid dir from line135 to find it;

In other word, on my understanding, mmpose use image_id from bbox_1000pic.json which read at line72--73(for example image_id:21, btw, person_keypoints_valid_20pic.json only has 20images, doesnot has image_id:21), then, try to find it in valid dir
(try to find 000 000 000 021.jpg in {data_root}/valid_21pic/). Error showed as: FileNotFoundError: img file does not exist: data/valid_20pic/000000000021.jpg

So~ config file can be only put 1 train_1000_bbox.json in it, but there are 2 ,train_1000.json and valid_20.json, in train_pipeline,valid_pipeline,
Are there any way to change config file to provide valid_20_bbox.json for valid_pipeline, or teach it to read bbox info from valid_20.json.

Thankyou very much.

configs/top_down/resnet/coco/res50_coco_256x192.py
missed
test_pipeline = valid_pipeline

where can i download this checkpoint。

and i use hrnet_w48_coco_256x192-b9e0b3ab_20200708.pth
but i get the error：
File "demo/top_down_video_demo_with_mmdet.py", line 118, in
main()
File "demo/top_down_video_demo_with_mmdet.py", line 91, in main
format='xyxy')
File "/home/cl/gitclone/mmpose/mmpose/apis/inference.py", line 228, in inference_pose_model
pose = _inference_single_pose_model(model, img_or_path, bbox)
File "/home/cl/gitclone/mmpose/mmpose/apis/inference.py", line 175, in _inference_single_pose_model
data = test_pipeline(data)
File "/home/cl/gitclone/mmpose/mmpose/datasets/pipelines/shared_transform.py", line 78, in call
data = t(data)
File "/home/cl/gitclone/mmpose/mmpose/datasets/pipelines/top_down_transform.py", line 181, in call
r = results['rotation']
KeyError: 'rotation'

Hi,

Thank you for your interesting work. I am trying to test HigherHRNet on a collection of videos I've found very challenging for existing human pose estimation models. However when I try to execute the following command:

python demo/video_demo_with_mmdet.py $SCRATCH/mmdetection/configs/faster_rcnn/faster_rcnn_r50_fpn_1x_coco.py $SCRATCH/mmdetection/checkpoints/faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth configs/bottom_up/higherhrnet/coco/higher_hrnet48_coco_512x512.py checkpoints/higher_hrnet48_coco_512x512-60fedcbc_20200712.pth --video-path $SCRATCH/VIBE/sample_video.mp4 --show --out-video-root ./

I get a number of config related issues, the first is that the image size seems to be treated as both an integer as well as a list, I made a temporary fix with a try except statement, but the second issue I'm dealing with now is the lack of a flip_index.

I am not sure what purpose the flip_index has but I've tried making it 0 or [0,1] as I saw in another example and neither of these work.

For a detailed traceback I get

Traceback (most recent call last):
  File "demo/video_demo_with_mmdet.py", line 117, in <module>
    main()
  File "demo/video_demo_with_mmdet.py", line 90, in main
    format='xyxy')
  File "/mmpose/mmpose/apis/inference.py", line 238, in inference_pose_model
    pose = _inference_single_pose_model(model, img_or_path, bbox)
  File "mmpose/mmpose/apis/inference.py", line 194, in _inference_single_pose_model
    return_loss=False, img=data['img'], img_metas=data['img_metas'])
  File ".conda/envs/pytorch150/lib/python3.7/site-packages/torch/nn/modules/module.py", line 550, in __call__
    result = self.forward(*input, **kwargs)
  File "mmpose/mmpose/models/detectors/bottom_up.py", line 103, in forward
    return self.forward_test(img, img_metas, **kwargs)
  File "mmpose/mmpose/models/detectors/bottom_up.py", line 218, in forward_test
    self.test_cfg['flip_test'])
  File "mmpose/mmpose/core/evaluation/bottom_up_eval.py", line 157, in aggregate_results
    heatmaps[1]) / 2.0 if flip_test else heatmaps[0]

Hope you can help me, I'm really excited about your work and hoping to be able to make contributions in the future.

Hello and first of all thanks for the great tool you have built.

I wanted to ask some questions, I run the pretrained model on a 1080 ti on a realtime video and obtain so-so performances :
res50_coco_256x192: 12 FPS
mobilenetv2_coco_256x192: 15 FPS

Am I missing something or is it just that the GPU isn't strong enough. Thanks!

a global search & replace of https://openmmlab.oss-accelerate.aliyuncs.com is ok

hello, I've read the evaluation code for MPII dataset, it calculated about the average accracy for all the keypoints. However, on its official website, the evaluation has been divided into two parts: single-person and multi-person, a little bit different from this repo. So does this repo use the same evaluaion tools as the official realization or only calculate the averaged results for all keypoints in all images? Hope for your reply! Thanks a lot!!

Pardon me, buddies, I'm planed to training the wonderful frame, and used this command: ./tools/dist_train.sh configs/top_down/resnet/coco/res50_coco_256x192.py 2 --no-validate;
So, I got a cuda memory problem: when I set gpu_num=2, it's return a cuda memory error. If I changed gpu_num =1, it's works, and begin training, no memory error return.
The Nvidia configuration (set gpu=1, it has been running):
| NVIDIA-SMI 440.36 Driver Version: 440.36 CUDA Version: 10.2 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 GeForce RTX 2070 Off | 00000000:01:00.0 On | N/A |
| 90% 71C P2 170W / 175W | 7966MiB / 7979MiB | 97% Default |
+-------------------------------+----------------------+----------------------+
| 1 GeForce RTX 2070 Off | 00000000:03:00.0 Off | N/A |
| 50% 40C P8 9W / 175W | 12MiB / 7982MiB | 0% Default |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage

Did my torch envs problem? Or system problem?

So I was trying to test the mpii dataset using both test.py and ./tool/dist_test.sh and came across this small issue. It looks like mpii uses the evaluation metric PCKh rather than mPA so I changed the args as follows

For test.py

 python tools/test.py configs/top_down/hrnet/mpii/hrnet_w32_mpii_256x256.py ~/0Dev/data/models/hrnet_w32_mpii_256x256-6c4f923f_20200812.pth --eval PCKh

for ./tools/dist_test.sh

./tools/dist_test.sh configs/top_down/hrnet/mpii/hrnet_w32_mpii_256x256.py ~/0Dev/data/models/hrnet_w32_mpii_256x256-6c4f923f_20200812.pth 2 --eval PCKh

I get the assertion error saying that metric != PCKh and modified the error so I could figure out what was going on and saw

assert metric == 'PCKh', f"got {metric}"
AssertionError: got ['PCKh']

I'm fairly new to this repo so I don't know the ins and outs enough to make a pull request, so I wanted to make you guys aware of this issue!

So I'm trying to learn how to use this repo, but I can't figure out how to save the model. I assumed it would be saved in the same directory as the logs are created but don't see any .pth files there.

I've done the following when training a resnet50 model on the mpii dataset on 2 gpus

./tools/dist_train.sh configs/top_down/resnet/mpii/res50_mpii_256x256_test.py 2

I have a pretrained model at models/pytorch/imagenet/resnet50-19c8e357.pth

I train for 2 epochs and get the .log and .log.json files + a pred.mat though I'm not sure what the .mat format is?
Where does the .pth file save? I've tried digging into the code but haven't really found where and it seems like currently, I can only finetune to save a model? does it only save based on a certain eval metric (PCKh/mAP)?

Any help would be appreciated, thanks!

Don't know if it is just me. Cloning the repository form the mentioned gitlab link doesn't work. The connection times out.

(HRNet) akilesh@heracleiapc:~/PyCharmProjects$ git clone [email protected]:open-mmlab/mmpose.git
Cloning into 'mmpose'...
ssh: connect to host gitlab.sz.sensetime.com port 22: Connection timed out
fatal: Could not read from remote repository.

Please make sure you have the correct access rightsand the repository exists.

I cloned with the github link though.!

It's a really fantastic work! Can you share the processing code on how to transform the .mat file to json file? I've found the annotations is in the type of .mat file on the MPII Dataset website. Thanks a lot!!

We keep this issue open to collect feature requests from users and hear your voice. Our monthly release plan is also available here.

You can either:

1. Suggest a new feature by leaving a comment.
2. Vote for a feature request with 👍 or be against with 👎. (Remember that developers are busy and cannot respond to all feature requests, so vote for your most favorable one!)
3. Tell us that you would like to help implement one of the features in the list or review the PRs. (This is the greatest things to hear about!)

Hello,
I would like to ask if it's possible to convert MMPose models to ONNX format for using with TensorRT for better real-time performance.
Thanks

cc. @yaochaorui

when ‘python setup.py develop’, no such file or directory:'mmpose/VERSION'

I got an error like title when I try demo using mmdetection.▼

https://mmpose.readthedocs.io/en/latest/getting_started.html#using-mmdet-for-human-bounding-box-detection

How can I fix it?

Does MMPose support a temporal analysis of videos yet, where the pose estimation is not done on a frame-by-frame basis, but rather using temporal information, like in this paper on VideoPose3D ?
Supporting temporal frameworks would need the datasets to consist of an unordered set of videos, where each video is a sequential ordered list of frames. Furthermore, the splitting of training and validation set, the definition of batches and the input to the model would also need to work accordingly.