that's ok

@hasanirtiza I solve the problem with this modification. I think it's could be a version problem. Thanks for your reply.

from pedestron.

Which dataset are you trying to train on ? Did you change anything in the configuration file ? Could you paste your config file here ?

from pedestron.

Thank you for your reply . i don not change any configuration without pretrained and path. And config file as follows:

model settings

model = dict(
type='CascadeRCNN',
num_stages=3,
pretrained=None,
backbone=dict(
type='HRNet',
extra=dict(
stage1=dict(
num_modules=1,
num_branches=1,
block='BOTTLENECK',
num_blocks=(4, ),
num_channels=(64, )),
stage2=dict(
num_modules=1,
num_branches=2,
block='BASIC',
num_blocks=(4, 4),
num_channels=(32, 64)),
stage3=dict(
num_modules=4,
num_branches=3,
block='BASIC',
num_blocks=(4, 4, 4),
num_channels=(32, 64, 128)),
stage4=dict(
num_modules=3,
num_branches=4,
block='BASIC',
num_blocks=(4, 4, 4, 4),
num_channels=(32, 64, 128, 256))
),
#frozen_stages=-1,
norm_eval=False,
),
neck=dict(
type='HRFPN',
in_channels=[32, 64, 128, 256],
out_channels=256),
rpn_head=dict(
type='RPNHead',
in_channels=256,
feat_channels=256,
anchor_scales=[8],
anchor_ratios=[0.5, 1.0, 2.0],
anchor_strides=[4, 8, 16, 32, 64],
target_means=[.0, .0, .0, .0],
target_stds=[1.0, 1.0, 1.0, 1.0],
loss_cls=dict(
type='CrossEntropyLoss', use_sigmoid=True, loss_weight=1.0),
loss_bbox=dict(type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=1.0)),
bbox_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(
type='RoIAlign',
out_size=7,
sample_num=2),
out_channels=256,
featmap_strides=[4, 8, 16, 32]),
bbox_head=[
dict(
type='SharedFCBBoxHead',
num_fcs=2,
in_channels=256,
fc_out_channels=1024,
roi_feat_size=7,
num_classes=2,
target_means=[0., 0., 0., 0.],
target_stds=[0.1, 0.1, 0.2, 0.2],
reg_class_agnostic=True,
loss_cls=dict(
type='CrossEntropyLoss',
use_sigmoid=False,
loss_weight=1.0),
loss_bbox=dict(
type='SmoothL1Loss',
beta=1.0,
loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
in_channels=256,
fc_out_channels=1024,
roi_feat_size=7,
num_classes=2,
target_means=[0., 0., 0., 0.],
target_stds=[0.05, 0.05, 0.1, 0.1],
reg_class_agnostic=True,
loss_cls=dict(
type='CrossEntropyLoss',
use_sigmoid=False,
loss_weight=1.0),
loss_bbox=dict(
type='SmoothL1Loss',
beta=1.0,
loss_weight=1.0)),
dict(
type='SharedFCBBoxHead',
num_fcs=2,
in_channels=256,
fc_out_channels=1024,
roi_feat_size=7,
num_classes=2,
target_means=[0., 0., 0., 0.],
target_stds=[0.033, 0.033, 0.067, 0.067],
reg_class_agnostic=True,
loss_cls=dict(
type='CrossEntropyLoss',
use_sigmoid=False,
loss_weight=1.0),
loss_bbox=dict(
type='SmoothL1Loss',
beta=1.0,
loss_weight=1.0)),
],
mask_roi_extractor=dict(
type='SingleRoIExtractor',
roi_layer=dict(type='RoIAlign', out_size=14, sample_num=2),
out_channels=256,
featmap_strides=[4, 8, 16, 32]),
mask_head=dict(
type='FCNMaskHead',
num_convs=4,
in_channels=256,
conv_out_channels=256,
num_classes=2,
loss_mask=dict(
type='CrossEntropyLoss', use_mask=True, loss_weight=1))
)

model training and testing settings

train_cfg = dict(
rpn=dict(
assigner=dict(
type='MaxIoUAssigner',
pos_iou_thr=0.7,
neg_iou_thr=0.3,
min_pos_iou=0.3,
ignore_iof_thr=0.7),
sampler=dict(
type='RandomSampler',
num=256,
pos_fraction=0.5,
neg_pos_ub=-1,
add_gt_as_proposals=False),
allowed_border=0,
pos_weight=-1,
debug=False),
rpn_proposal=dict(
nms_across_levels=False,
nms_pre=2000,
nms_post=2000,
max_num=2000,
nms_thr=0.7,
min_bbox_size=0),
rcnn=[
dict(
assigner=dict(
type='MaxIoUAssigner',
pos_iou_thr=0.5,
neg_iou_thr=0.5,
min_pos_iou=0.5,
ignore_iof_thr=0.7),
sampler=dict(
type='RandomSampler',
num=512,
pos_fraction=0.25,
neg_pos_ub=-1,
add_gt_as_proposals=True),
mask_size=28,
pos_weight=-1,
debug=False),
dict(
assigner=dict(
type='MaxIoUAssigner',
pos_iou_thr=0.6,
neg_iou_thr=0.6,
min_pos_iou=0.6,
ignore_iof_thr=0.7),
sampler=dict(
type='RandomSampler',
num=512,
pos_fraction=0.25,
neg_pos_ub=-1,
add_gt_as_proposals=True),
mask_size=28,
pos_weight=-1,
debug=False),
dict(
assigner=dict(
type='MaxIoUAssigner',
pos_iou_thr=0.7,
neg_iou_thr=0.7,
min_pos_iou=0.7,
ignore_iof_thr=0.7),
sampler=dict(
type='RandomSampler',
num=512,
pos_fraction=0.25,
neg_pos_ub=-1,
add_gt_as_proposals=True),
mask_size=28,
pos_weight=-1,
debug=False)
],
stage_loss_weights=[1, 0.5, 0.25])
test_cfg = dict(
rpn=dict(
nms_across_levels=False,
nms_pre=1000,
nms_post=1000,
max_num=1000,
nms_thr=0.7,
min_bbox_size=0),
rcnn=dict(
score_thr=0.001,
nms=dict(type='soft_nms', iou_thr=0.5),
max_per_img=100,
mask_thr_binary=0.5),
keep_all_stages=False)

dataset settings

dataset_type = 'CocoDataset'
data_root = 'datasets/CrowdHuman/'
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53],
std=[58.395, 57.12, 57.375],
to_rgb=True)
data = dict(
imgs_per_gpu=1,
workers_per_gpu=5,
train=dict(
type=dataset_type,
# ann_file=[data_root + 'train.json',
# data_root + 'val.json'],
# img_prefix=[data_root + 'Images',
# data_root + 'Images_val'],

ann_file=data_root + 'train.json',
img_prefix=data_root + 'Images',
img_scale=[(1216, 608),(2048, 1024)],
    multiscale_mode='range',
    img_norm_cfg=img_norm_cfg,
    size_divisor=32,
    flip_ratio=0.5,
    with_mask=True,
    with_crowd=True,
    with_label=True,
    extra_aug=dict(
        photo_metric_distortion=dict(brightness_delta=180, contrast_range=(0.5, 1.5),
             saturation_range=(0.5, 1.5), hue_delta=18),
         random_crop=dict(min_ious=(0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9), min_crop_size=0.1),
     ),
),
test=dict(
    type=dataset_type,
    ann_file=data_root + 'val.json',
    img_prefix=data_root + 'Images_val',
    img_scale=(2048, 1024),
    img_norm_cfg=img_norm_cfg,
    size_divisor=32,
    flip_ratio=0,
    with_mask=False,
    with_label=False,
    test_mode=True))

optimizer

mean_teacher=True
optimizer = dict(type='SGD', lr=0.02, momentum=0.9, weight_decay=0.0001)
optimizer_config = dict(grad_clip=dict(max_norm=35, norm_type=2), mean_teacher = dict(alpha=0.999))

learning policy

lr_config = dict(
policy='cosine',
warmup='linear',
warmup_iters=500,
warmup_ratio=1.0 / 3,
step=[110, 160])
checkpoint_config = dict(interval=1)

yapf:disable

log_config = dict(
interval=50,
hooks=[
dict(type='TextLoggerHook'),
# dict(type='TensorboardLoggerHook')
])

yapf:enable

runtime settings

total_epochs = 240
dist_params = dict(backend='nccl')
log_level = 'INFO'
work_dir = './work_dirs/cityperson_cascade_rcnn_hrnetv2p_w32'
load_from = None
resume_from = None
workflow = [('train', 1)]

from pedestron.

Config seems fine and I cannot reproduce this issue locally. Now can you please paste your PyTorch etc. versions ?

from pedestron.

My env as follows:

This file may be used to create an environment using:

$ conda create --name --file

platform: linux-64

_libgcc_mutex=0.1=main
_pytorch_select=0.2=gpu_0
blas=1.0=mkl
ca-certificates=2020.1.1=0
certifi=2020.4.5.1=py37_0
cffi=1.14.0=py37h2e261b9_0
cudatoolkit=10.0.130=0
cudnn=7.6.5=cuda10.0_0
cycler=0.10.0=py37_0
cython=0.29.17=py37he6710b0_0
freetype=2.9.1=h8a8886c_1
intel-openmp=2020.0=166
jpeg=9b=h024ee3a_2
libedit=3.1.20181209=hc058e9b_0
libffi=3.2.1=hd88cf55_4
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_0
mkl=2020.0=166
mkl-service=2.3.0=py37he904b0f_0
mkl_fft=1.0.15=py37ha843d7b_0
mkl_random=1.1.0=py37hd6b4f25_0
ncurses=6.2=he6710b0_1
ninja=1.9.0=py37hfd86e86_0
numpy=1.18.1=py37h4f9e942_0
numpy-base=1.18.1=py37hde5b4d6_1
olefile=0.46=py_0
openssl=1.0.2u=h7b6447c_0
pillow=7.1.2=py37hb39fc2d_0
pip=20.0.2=py37_1
pycparser=2.20=py_0
pyparsing=2.4.7=py_0
python=3.7.0=h6e4f718_3
pytorch=1.2.0=cuda100py37h938c94c_0
readline=7.0=h7b6447c_5
setuptools=46.1.3=py37_0
six=1.14.0=py37_0
sqlite=3.31.1=h62c20be_1
tk=8.6.8=hbc83047_0
torchvision=0.4.0=cuda100py37hecfc37a_0
wheel=0.34.2=py37_0
xz=5.2.5=h7b6447c_0
zlib=1.2.11=h7b6447c_3
zstd=1.3.7=h0b5b093_0

from pedestron.

Can you run the demo.py (using any of the pre-trained model) ?

from pedestron.

I can run the demo.py (use crowdhuman pre-trained model). Only l want to train on crowdhuman from scratch.

from pedestron.

I cannot reproduce the error. I have just cloned the repo and tried training on crowdhuman from the scratch. It is running fine on my system (500 iterations so far). So, it can be either an issue in the installation (did it compile normally after running python setup.py develop ?) or there is some incompatibility issue between the libs etc. One more thing as a caveat, in case you are running mmdetection framework as well for some other projects, make sure you have a separate environment for Pedestron. Since, mmdetection has been upgraded to a newer version (mmdet and mmcv), at least in my case I faced some issues.

from pedestron.

Thank you , there is no error in the compilation.
l have created an independent conda environment.
I don not run mmmdetection framework on another project.
Can you provide a full environment file?
And i have a question that the num_classes are set to 2 in the config file. Why not set to 3 (person, mask and background)?

from pedestron.

# packages in environment at envs/Pedestron:
#
# Name                    Version                   Build  Channel
_libgcc_mutex             0.1                        main  
addict                    2.2.1                    pypi_0    pypi
attrdict                  2.0.1                    pypi_0    pypi
blas                      1.0                         mkl  
bzip2                     1.0.6                h9a117a8_4  
ca-certificates           2019.11.27                    0  
caffe2-cuda8.0-cudnn7     0.8.dev         py36_2018.05.14    caffe2
cairo                     1.14.12              h7636065_2  
certifi                   2019.11.28               py36_0  
cffi                      1.11.5           py36h9745a5d_0  
chardet                   3.0.4                    pypi_0    pypi
cudatoolkit               9.0                  h13b8566_0  
cycler                    0.10.0           py36h93f1223_0  
cython                    0.29.13                  pypi_0    pypi
dbus                      1.13.2               h714fa37_1  
decorator                 4.4.1                    pypi_0    pypi
expat                     2.2.5                he0dffb1_0  
ffmpeg                    3.4                  h7264315_0  
fontconfig                2.12.6               h49f89f6_0  
freeglut                  2.8.1                         0    https://repo.continuum.io/pkgs/free
freetype                  2.8                  hab7d2ae_1  
future                    0.16.0                   py36_1    https://repo.continuum.io/pkgs/free
gflags                    2.2.1                hf484d3e_0  
glib                      2.56.1               h000015b_0  
glog                      0.3.5                hf484d3e_1  
graphite2                 1.3.11               hf63cedd_1  
gst-plugins-base          1.14.0               hbbd80ab_1  
gstreamer                 1.14.0               hb453b48_1  
h5py                      2.8.0            py36hca9c191_0  
harfbuzz                  1.7.6                h5f0a787_1  
hdf5                      1.8.18               h6792536_1  
icu                       58.2                 h9c2bf20_1  
idna                      2.8                      pypi_0    pypi
imageio                   2.6.1                    pypi_0    pypi
intel-openmp              2018.0.0                      8  
jasper                    2.0.14               h07fcdf6_0  
jpeg                      9b                   h024ee3a_2  
kiwisolver                1.0.1            py36h764f252_0  
libedit                   3.1                  heed3624_0  
libffi                    3.2.1                hd88cf55_4  
libgcc-ng                 7.2.0                hdf63c60_3  
libgfortran-ng            7.2.0                hdf63c60_3  
libglu                    9.0.0                h0c0bdc1_1  
libopus                   1.2.1                hb9ed12e_0  
libpng                    1.6.34               hb9fc6fc_0  
libprotobuf               3.5.2                h6f1eeef_0  
libstdcxx-ng              7.2.0                hdf63c60_3  
libtiff                   4.0.9                h28f6b97_0  
libvpx                    1.6.1                h888fd40_0  
libxcb                    1.13                 h1bed415_1  
libxml2                   2.9.8                hf84eae3_0  
matplotlib                2.2.2            py36h0e671d2_1  
mkl                       2018.0.2                      1  
mkl_fft                   1.0.1            py36h3010b51_0  
mkl_random                1.0.1            py36h629b387_0  
mmcv                      0.2.14                   pypi_0    pypi
mmdet                     0.6.0+47c8e78             dev_0    <develop>
ncurses                   6.0                  h9df7e31_2  
networkx                  2.4                      pypi_0    pypi
ninja                     1.8.2            py36h6bb024c_1  
numpy                     1.14.3           py36hcd700cb_1  
numpy-base                1.14.3           py36h9be14a7_1  
olefile                   0.45.1                   py36_0  
opencv                    3.3.1            py36h9248ab4_2  
opencv-python             4.1.1.26                 pypi_0    pypi
openssl                   1.0.2p               h14c3975_0  
pandas                    0.23.4           py36h04863e7_0  
patsy                     0.5.1                    pypi_0    pypi
pcre                      8.42                 h439df22_0  
pillow                    5.1.0            py36h3deb7b8_0  
pip                       19.2.3                   pypi_0    pypi
pixman                    0.34.0               hceecf20_3  
protobuf                  3.5.2                    py36_0    conda-forge
psutil                    5.4.6            py36h14c3975_0    anaconda
pycocotools               2.0                      pypi_0    pypi
pycparser                 2.18             py36hf9f622e_1  
pyparsing                 2.2.0            py36hee85983_1  
pyqt                      5.9.2            py36h751905a_0  
python                    3.6.5                hc3d631a_2  
python-dateutil           2.7.2                    py36_0  
pytorch                   1.1.0           py3.6_cuda9.0.176_cudnn7.5.1_0    pytorch
pytz                      2018.4                   py36_0  
pywavelets                1.1.1                    pypi_0    pypi
pyyaml                    3.12             py36hafb9ca4_1  
qt                        5.9.5                h7e424d6_0  
readline                  7.0                  ha6073c6_4  
requests                  2.22.0                   pypi_0    pypi
scikit-image              0.16.2                   pypi_0    pypi
scikit-learn              0.19.1           py36h7aa7ec6_0  
scipy                     1.1.0            py36hfc37229_0  
seaborn                   0.9.0              pyh91ea838_1  
setuptools                39.1.0                   py36_0  
sip                       4.19.8           py36hf484d3e_0  
six                       1.11.0           py36h372c433_1  
sqlite                    3.23.1               he433501_0  
statsmodels               0.10.1                   pypi_0    pypi
tbb                       2018.0.4             h6bb024c_1    anaconda
tbb4py                    2018.0.4         py36h6bb024c_1    anaconda
tensorboardx              1.4                      pypi_0    pypi
tk                        8.6.7                hc745277_3  
torchvision               0.3.0           py36_cu9.0.176_1    pytorch
tornado                   5.0.2                    py36_0  
tqdm                      4.23.0                   py36_0  
urllib3                   1.25.7                   pypi_0    pypi
wheel                     0.31.0                   py36_0  
xz                        5.2.3                h5e939de_4  
yaml                      0.1.7                had09818_2  
zlib                      1.2.11               ha838bed_2

Lots of unwanted packages as well, ignore them.

from pedestron.

Thank you , there is no error in the compilation.
l have created an independent conda environment.
I don not run mmmdetection framework on another project.
Can you provide a full environment file?
And i have a question that the num_classes are set to 2 in the config file. Why not set to 3 (person, mask and background)?

For the reason, Pedestron generates mask on the fly #13. Therefore, we are not really interested in the masks. You can change it if you want.

from pedestron.

Thank you, I have copied your environment and it has run successfully.
I want to know how to load your pre-training model and let it continue to train on crowdhuman dataset?
And can you tell me how much the MAP on crowdhuman dataset?

from pedestron.

In the config file, you have load_from = None. Change None to the model you want to load from, for example,load_from = 'my_path/epoch_1.pth'. I think the published model has an AP around ~84.

from pedestron.

When I try to load your pre-trained model (epoch_19.pth.stu), it reported an error.
As follows:
OSError: epoch_19.pth is not a checkpoint file

from pedestron.

Yes, you actually need a teacher model not a student one (not .stu but .pth). Can you try simply making a copy of epoch_19.pth.stu and name it epoch_19.pth and pass this .pth model. See if it works.

from pedestron.

You are right!
Thank you for your patient reply!

from pedestron.

that's ok

from pedestron.

@hasanirtiza

what is the difference between teacher model and student model?

from pedestron.

In both models weights are saved slightly differently. It is a method that does some sort of smoothing of weights during training and they are helpful. You can read more about it in this NIPS paper

from pedestron.

What is the cause of this problem? How did you solve it? @muzishen

from pedestron.

According the arthor reply,Yes, you actually need a teacher model not a student one (not .stu but .pth). Can you try simply making a copy of epoch_19.pth.stu and name it epoch_19.pth and pass this .pth model. See if it works.

from pedestron.

Sorry, I don't think I made myself clear. I mean this problem ’RuntimeError: Expected object of scalar type Byte but got scalar type Bool for argument #2 'other'‘ @muzishen

from pedestron.

@yuanvq Actually this is an error that I have also reproduced locally. Most probably it is down to the environment (mmdet, mmcv torch etc), make sure versions are as suggested by the repo. Another issue is that, if you are also running mmdetection (newer versions) or any other repo that uses mmdetection and you do not have separate environments, you might also encounter this problem.

By the way did you manage to solve it ?

from pedestron.