HashNet.py

from utils.tools import *
from network import *

import os
import torch
import torch.optim as optim
import time
import numpy as np

torch.multiprocessing.set_sharing_strategy('file_system')


# HashNet(ICCV2017)
# paper [HashNet: Deep Learning to Hash by Continuation](http://openaccess.thecvf.com/content_ICCV_2017/papers/Cao_HashNet_Deep_Learning_ICCV_2017_paper.pdf)
# code [HashNet caffe and pytorch](https://github.com/thuml/HashNet)

def get_config():
    config = {
        "alpha": 0.1,
        # "optimizer":{"type":  optim.SGD, "optim_params": {"lr": 0.001, "weight_decay": 10 ** -5}, "lr_type": "step"},
        "optimizer": {"type": optim.RMSprop, "optim_params": {"lr": 1e-5, "weight_decay": 10 ** -5}, "lr_type": "step"},
        "info": "[HashNet]",
        "step_continuation": 20,
        "resize_size": 256,
        "crop_size": 224,
        "batch_size": 64,
        "net": AlexNet,
        # "net":ResNet,
        # "dataset": "cifar10",
        "dataset": "cifar10-1",
        # "dataset": "cifar10-2",
        # "dataset": "coco",
        # "dataset": "mirflickr",
        # "dataset": "voc2012",
        # "dataset": "imagenet",
        # "dataset": "nuswide_21",
        # "dataset": "nuswide_21_m",
        # "dataset": "nuswide_81_m",
        "epoch": 150,
        "test_map": 15,
        "save_path": "save/HashNet",
        # "device":torch.device("cpu"),
        "device": torch.device("cuda:0"),
        "bit_list": [48],
    }
    config = config_dataset(config)
    return config


class HashNetLoss(torch.nn.Module):
    def __init__(self, config, bit):
        super(HashNetLoss, self).__init__()
        self.scale = 1

    def forward(self, u, y, ind, config):
        u = torch.tanh(self.scale * u)
        S = (y @ y.t() > 0).float()
        sigmoid_alpha = config["alpha"]
        dot_product = sigmoid_alpha * u @ u.t()
        mask_positive = S > 0
        mask_negative = (1 - S).bool()
        
        neg_log_probe = dot_product + torch.log(1 + torch.exp(-dot_product)) -  S * dot_product
        S1 = torch.sum(mask_positive.float())
        S0 = torch.sum(mask_negative.float())
        S = S0 + S1

        neg_log_probe[mask_positive] = neg_log_probe[mask_positive] * S / S1
        neg_log_probe[mask_negative] = neg_log_probe[mask_negative] * S / S0

        loss = torch.sum(neg_log_probe) / S
        return loss


def train_val(config, bit):
    device = config["device"]
    train_loader, test_loader, dataset_loader, num_train, num_test, num_dataset = get_data(config)
    config["num_train"] = num_train
    net = config["net"](bit).to(device)

    optimizer = config["optimizer"]["type"](net.parameters(), **(config["optimizer"]["optim_params"]))

    criterion = HashNetLoss(config, bit)

    Best_mAP = 0

    for epoch in range(config["epoch"]):
        criterion.scale = (epoch // config["step_continuation"] + 1) ** 0.5

        current_time = time.strftime('%H:%M:%S', time.localtime(time.time()))

        print("%s[%2d/%2d][%s] bit:%d, dataset:%s, scale:%.3f, training...." % (
            config["info"], epoch + 1, config["epoch"], current_time, bit, config["dataset"], criterion.scale), end="")

        net.train()

        train_loss = 0
        for image, label, ind in train_loader:

            image = image.to(device)
            label = label.to(device)

            optimizer.zero_grad()
            u = net(image)

            loss = criterion(u, label.float(), ind, config)
            train_loss += loss.item()

            loss.backward()
            optimizer.step()

        train_loss = train_loss / len(train_loader)

        print("\b\b\b\b\b\b\b loss:%.3f" % (train_loss))

        if (epoch + 1) % config["test_map"] == 0:
            Best_mAP = validate(config, Best_mAP, test_loader, dataset_loader, net, bit, epoch, num_dataset)


if __name__ == "__main__":
    config = get_config()
    print(config)
    for bit in config["bit_list"]:
        config["pr_curve_path"] = f"log/alexnet/HashNet_{config['dataset']}_{bit}.json"
        train_val(config, bit)