extract_features.py

# --------------------------------------------------------
# Swin Transformer
# Copyright (c) 2021 Microsoft
# Licensed under The MIT License [see LICENSE for details]
# Written by Ze Liu
# --------------------------------------------------------

import os
import time
import json
import random
import argparse
import datetime
import numpy as np
from torchvision import transforms

import torch
import torch.backends.cudnn as cudnn
import torch.distributed as dist

from timm.loss import LabelSmoothingCrossEntropy, SoftTargetCrossEntropy
from timm.utils import accuracy, AverageMeter
from PIL import Image

from models import build_model
from data import build_loader
from lr_scheduler import build_scheduler
from optimizer import build_optimizer
from logger import create_logger
from utils import load_checkpoint, load_pretrained, save_checkpoint, NativeScalerWithGradNormCount, auto_resume_helper, \
    reduce_tensor, parse_option, con_loss, instance_con_loss
from pytorch_grad_cam import GradCAM
from pytorch_grad_cam.utils.model_targets import ClassifierOutputTarget
from pytorch_grad_cam.utils.image import show_cam_on_image
import matplotlib.pyplot as plt
from tqdm import tqdm

def reshape_transform(tensor, height=14, width=14):
    result = tensor.reshape(tensor.size(0),
        height, width, tensor.size(2))

    # Bring the channels to the first dimension,
    # like in CNNs.
    result = result.transpose(2, 3).transpose(1, 2)
    return result


@torch.no_grad()
def main(config):
    config.defrost()
    data_loader_train, config.MODEL.NUM_CLASSES = build_loader(config, logger=logger, is_pretrain=False, is_train=True)
    config.freeze()
    data_loader_val, _ = build_loader(config, logger=logger, is_pretrain=False, is_train=False)
    mixup_fn = None

    logger.info(f"Creating model:{config.MODEL.TYPE}/{config.MODEL.NAME}")
    model = build_model(config)
    logger.info(str(model))

    n_parameters = sum(p.numel() for p in model.parameters() if p.requires_grad)
    logger.info(f"number of params: {n_parameters}")
    if hasattr(model, 'flops'):
        flops = model.flops()
        logger.info(f"number of GFLOPs: {flops / 1e9}")

    model.cuda()
    model_without_ddp = model

    optimizer = build_optimizer(config, model)
    model = torch.nn.parallel.DistributedDataParallel(model, device_ids=[config.LOCAL_RANK], broadcast_buffers=False)
    loss_scaler = NativeScalerWithGradNormCount()

    if config.TRAIN.ACCUMULATION_STEPS > 1:
        lr_scheduler = build_scheduler(config, optimizer, len(data_loader_train) // config.TRAIN.ACCUMULATION_STEPS)
    else:
        lr_scheduler = build_scheduler(config, optimizer, len(data_loader_train))

    if config.AUG.MIXUP > 0.:
        # smoothing is handled with mixup label transform
        criterion = SoftTargetCrossEntropy()
    elif config.MODEL.LABEL_SMOOTHING > 0.:
        criterion = LabelSmoothingCrossEntropy(smoothing=config.MODEL.LABEL_SMOOTHING)
    else:
        criterion = torch.nn.CrossEntropyLoss()

    max_accuracy = 0.0

    if config.MODEL.PRETRAINED and (not config.MODEL.RESUME):
        load_pretrained(config, model_without_ddp, logger)
        # acc1, acc5, loss = validate(config, data_loader_val, model)
        # logger.info(f"Accuracy of the network on the test images: {acc1:.2f}%")


    logger.info("Start training")
    start_time = time.time()
    best_epoch = 0
    target_layers = [model.module.layers[-1].blocks[-1].norm1]
    cam = GradCAM(model=model, target_layers=target_layers, use_cuda=True,reshape_transform=reshape_transform)
    root = os.path.join('./visualization', config.DATA.DATASET)
    record = {}
    for i, data in tqdm(enumerate(data_loader_val)):
        images = data[0].cuda(non_blocking=True)
        labels = data[1]
        img_names = data[2]
        _, features = model(images)
        record[img_names[0]] = {'feature': features.detach().cpu().numpy(),'label': labels[0]}
    torch.save(record,'./output/records/soygbl/'+config.DATA.DATASET+'_our.pth')



@torch.no_grad()
def validate(config, data_loader, model):
    criterion = torch.nn.CrossEntropyLoss()
    model.eval()

    batch_time = AverageMeter()
    loss_meter = AverageMeter()
    acc1_meter = AverageMeter()
    acc5_meter = AverageMeter()

    end = time.time()
    for idx, batch in enumerate(data_loader):
        images = batch[0].cuda(non_blocking=True)
        labels = batch[-1].cuda(non_blocking=True)
        # images = images.cuda(non_blocking=True)
        # labels = labels.cuda(non_blocking=True)

        # compute output
        with torch.cuda.amp.autocast(enabled=config.AMP_ENABLE):
            output, _ = model(images)

        # measure accuracy and record loss
        loss = criterion(output, labels)
        acc1, acc5 = accuracy(output, labels, topk=(1, 5))

        acc1 = reduce_tensor(acc1)
        acc5 = reduce_tensor(acc5)
        loss = reduce_tensor(loss)

        loss_meter.update(loss.item(), labels.size(0))
        acc1_meter.update(acc1.item(), labels.size(0))
        acc5_meter.update(acc5.item(), labels.size(0))

        # measure elapsed time
        batch_time.update(time.time() - end)
        end = time.time()

        if idx % config.PRINT_FREQ == 0:
            memory_used = torch.cuda.max_memory_allocated() / (1024.0 * 1024.0)
            logger.info(
                f'Test: [{idx}/{len(data_loader)}]\t'
                f'Time {batch_time.val:.3f} ({batch_time.avg:.3f})\t'
                f'Loss {loss_meter.val:.4f} ({loss_meter.avg:.4f})\t'
                f'Acc@1 {acc1_meter.val:.3f} ({acc1_meter.avg:.3f})\t'
                f'Acc@5 {acc5_meter.val:.3f} ({acc5_meter.avg:.3f})\t'
                f'Mem {memory_used:.0f}MB')
    logger.info(f' * Acc@1 {acc1_meter.avg:.3f} Acc@5 {acc5_meter.avg:.3f}')
    return acc1_meter.avg, acc5_meter.avg, loss_meter.avg


@torch.no_grad()
def throughput(data_loader, model, logger):
    model.eval()

    for idx, (images, _) in enumerate(data_loader):
        images = images.cuda(non_blocking=True)
        batch_size = images.shape[0]
        for i in range(50):
            model(images)
        torch.cuda.synchronize()
        logger.info(f"throughput averaged with 30 times")
        tic1 = time.time()
        for i in range(30):
            model(images)
        torch.cuda.synchronize()
        tic2 = time.time()
        logger.info(f"batch_size {batch_size} throughput {30 * batch_size / (tic2 - tic1)}")
        return


if __name__ == '__main__':
    args, config = parse_option()

    if config.AMP_OPT_LEVEL:
        print("[warning] Apex amp has been deprecated, please use pytorch amp instead!")

    if 'RANK' in os.environ and 'WORLD_SIZE' in os.environ:
        rank = int(os.environ["RANK"])
        world_size = int(os.environ['WORLD_SIZE'])
        print(f"RANK and WORLD_SIZE in environ: {rank}/{world_size}")
    else:
        rank = -1
        world_size = -1
    torch.cuda.set_device(config.LOCAL_RANK)
    torch.distributed.init_process_group(backend='nccl', init_method='env://', world_size=world_size, rank=rank)
    torch.distributed.barrier()

    seed = config.SEED + dist.get_rank()
    torch.manual_seed(seed)
    torch.cuda.manual_seed(seed)
    np.random.seed(seed)
    random.seed(seed)
    cudnn.benchmark = True

    # linear scale the learning rate according to total batch size, may not be optimal
    linear_scaled_lr = config.TRAIN.BASE_LR * config.DATA.BATCH_SIZE * dist.get_world_size() / 512.0
    linear_scaled_warmup_lr = config.TRAIN.WARMUP_LR * config.DATA.BATCH_SIZE * dist.get_world_size() / 512.0
    linear_scaled_min_lr = config.TRAIN.MIN_LR * config.DATA.BATCH_SIZE * dist.get_world_size() / 512.0
    # gradient accumulation also need to scale the learning rate
    if config.TRAIN.ACCUMULATION_STEPS > 1:
        linear_scaled_lr = linear_scaled_lr * config.TRAIN.ACCUMULATION_STEPS
        linear_scaled_warmup_lr = linear_scaled_warmup_lr * config.TRAIN.ACCUMULATION_STEPS
        linear_scaled_min_lr = linear_scaled_min_lr * config.TRAIN.ACCUMULATION_STEPS
    config.defrost()
    config.TRAIN.BASE_LR = linear_scaled_lr
    config.TRAIN.WARMUP_LR = linear_scaled_warmup_lr
    config.TRAIN.MIN_LR = linear_scaled_min_lr
    config.freeze()

    os.makedirs(config.OUTPUT, exist_ok=True)
    logger = create_logger(output_dir=config.OUTPUT, dist_rank=dist.get_rank(), name=f"{config.MODEL.NAME}")

    if dist.get_rank() == 0:
        path = os.path.join(config.OUTPUT, "config.json")
        with open(path, "w") as f:
            f.write(config.dump())
        logger.info(f"Full config saved to {path}")

    # print config
    logger.info(config.dump())
    logger.info(json.dumps(vars(args)))

    main(config)