Dear authors,
Hi, I am interested in your work. And I have noticed that you used JHU-CROWD++ as your evaluation dataset of crowd counting task. However, I did not find any code related to the preprocessing or loader of this dataset, thus I am wondering what are the procedures to deal with JHU-CROWD++.
I would greatly appreciate that if you could update the codes related to JHU-CROWD++ dataset.

Thanks in advance.

Hello, the author and everyone, I want to konw when the pytorch version will be completed, thanks very much.

In models/backbone/trainer.py, there occurs a problem, I can't find the CrowdCounter class.

import numpy as np

import torch
from torch import optim
from torch.autograd import Variable
from torch.optim.lr_scheduler import StepLR
from config import cfg
from misc.utils import *
import pdb
class Trainer():
    def __init__(self, dataloader, cfg_data, pwd):

        self.cfg_data = cfg_data

        self.data_mode = cfg.DATASET
        self.exp_name = cfg.EXP_NAME
        self.exp_path = cfg.EXP_PATH
        self.pwd = pwd

        self.net_name = cfg.NET
        self.net = CrowdCounter(cfg.GPU_ID,self.net_name).cuda()   # Here?!
        self.optimizer = optim.Adam(self.net.CCN.parameters(), lr=cfg.LR, weight_decay=1e-4)
        # self.optimizer = optim.SGD(self.net.parameters(), cfg.LR, momentum=0.95,weight_decay=5e-4)
        self.scheduler = StepLR(self.optimizer, step_size=cfg.NUM_EPOCH_LR_DECAY, gamma=cfg.LR_DECAY)

This is excellent work!

Could you please provide the training configurations for ResNet50 on the SHTech-PartA and JHU-CROWD++ datasets? Thank you in advance for your assistance!

Hello authors @zhiqic @zhiqicheng,

May I check how do we choose the data pre-processing step, in the paper there are two references made: one is based on .mat and .csv from C3 framework and one more repository [62] which uses .npy (referred from there to Bayesian repo which uses .npy as well)

Thank you much in advance for the suggestions.

Some Questions About How to Train this repo

I have been having some errors with the following files:

• models/backbone/train.py

• models/backbone/trainer.py

• config.py

  • on models/backbone/train.py :
    what is the "seed" expression between lines 8 and 11 written for?
    On line 43, when the code enters models/backbone/trainer.py, I get an "NameError: name "CrowdCounter" is not defined" error.

  • on models/backbone/trainer.py :
    the reason I am getting the name error in the models/backbone/train.py is the CrowdCounter function written on line 24 in this file, but it actually does not exist anywhere. Or at least I couldn`t find it.

    I cannot import cfg_data from the dataset.WE.setting specified on line 159. Because there is no such file in the specified dataset folder. Could the specified library be something other than WE? (such as SSHA, SSHB, UCF50 etc)

  • on config.py :
    On line 19: __C.NET = "Res50_SFCN" # net selection MCNN, Res50, CSRNet, SANet is written but I get an error because of the selected net, which is Res50_SFCN. Is it a problem if I use the network that I choose, not the default one?

Thanks in advance for all of your answers!

By the way, Although I have only just begun to interested in crowd counting analysis as a 3rd year student of electronics and communication engineering, I really enjoyed your paper and repository. 👍🏼

I'm having trouble understanding how counting papers evaluate and tune on various datasets. For the ShanghaitechB dataset, there's no given validation set. I'm having trouble following your code, but it seems you are tracking metrics on the test set during training and then saving that model. Are you selecting the best performance directly from the test set during training, or how are you determining the epoch where you will stop? Thank you and amazing paper!