Skip to main content

package for crowd counting

Project description

Crowd Counting Package

PyPi Version GitHub stars PyPi downloads

crowdcount is a library for crowd counting with Pytorch and supported by Fudan-VTS Research



  • pip install crowdcount --user --upgrade


Crowd counting task:

  • estimate the number of crowd
  • crowd counting demo

User guide:

  • models

     from crowdcount.models import * 
     # crowd counting models includes csr_net, mcnn, resnet50, resnet101, unet, vgg
  • transforms

     import crowdcount.transforms as cc_transforms
     # transforms
  • data_loader

     from import *
     # includes ShanghaiTech, UCF_QNRF, UCF_CC_50, Fudan-ShanghaiTech temporarily
  • data_preprocess

     from import *
     # gaussian preprocess for datasets
  • utils

     from crowdcount.utils import *
     # includes loss functions, optimizers, tensorboard and save function
  • engine

     from crowdcount.engine import train
     # start to train
     train(*args, **kwargs)
  • More details in document


from crowdcount.engine import train
from crowdcount.models import Res101
from import *
from crowdcount.utils import *
import crowdcount.transforms as cc_transforms
import torchvision.transforms as transforms

# init model
model = Res101()
# init transforms
img_transform = transforms.Compose([transforms.ToTensor(),
                                    transforms.Normalize(mean=[0.452016860247, 0.447249650955, 0.431981861591],
                                                         std=[0.23242045939, 0.224925786257, 0.221840232611])
gt_transform = cc_transforms.LabelEnlarge()
both_transform = cc_transforms.ComplexCompose([cc_transforms.TransposeFlip()])
# init dataset
train_set = ShanghaiTechDataset(mode="train",
test_set = ShanghaiTechDataset(mode="test",
# init loss
train_loss = AVGLoss()
test_loss = EnlargeLoss(100)
# init save function
saver = Saver(path="../exp/2019-12-22-main_sigma15_6")
# init tensorboard
tb = TensorBoard(path="../runs/2019-12-22-main_sigma15_6")
# start to train
train(model, train_set, test_set, train_loss, test_loss, optim="Adam", saver=saver, cuda_num=[3], train_batch=2,
      test_batch=2, learning_rate=1e-5, epoch_num=500, enlarge_num=100, tensorboard=tb)
  • you can find more demos in demo


we will add the results soon

Thanks for the supports from

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

crowdcount-0.1.1.tar.gz (16.7 kB view hashes)

Uploaded source

Built Distribution

crowdcount-0.1.1-py3-none-any.whl (30.8 kB view hashes)

Uploaded py3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page