FSC_finetune_cross.py

import argparse
import datetime
import json
import numpy as np
import os
import time
import random
from pathlib import Path
import math
import sys
from PIL import Image
import torch.nn as nn
from torchvision import transforms

import torch
import torch.backends.cudnn as cudnn
# from torch.utils.tensorboard import SummaryWriter
import scipy.ndimage as ndimage
from torch.utils.data import Dataset
import torchvision
import wandb

import timm.optim.optim_factory as optim_factory
import torchvision.transforms.functional as TF

import util.misc as misc
from util.misc import NativeScalerWithGradNormCount as NativeScaler
import util.lr_sched as lr_sched
from util.FSC147 import transform_train
import models_mae_cross

# import warnings
# warnings.filterwarnings('ignore')

torch.set_float32_matmul_precision('high')

def get_args_parser():
    parser = argparse.ArgumentParser('MAE pre-training', add_help=False)
    parser.add_argument('--batch_size', default=26, type=int,
                        help='Batch size per GPU (effective batch size is batch_size * accum_iter * # gpus')
    parser.add_argument('--epochs', default=200, type=int)
    parser.add_argument('--accum_iter', default=1, type=int,
                        help='Accumulate gradient iterations (for increasing the effective batch size under memory constraints)')

    # Model parameters
    parser.add_argument('--model', default='mae_vit_base_patch16', type=str, metavar='MODEL',
                        help='Name of model to train')

    parser.add_argument('--mask_ratio', default=0.5, type=float,
                        help='Masking ratio (percentage of removed patches).')

    parser.add_argument('--norm_pix_loss', action='store_true',
                        help='Use (per-patch) normalized pixels as targets for computing loss')
    parser.set_defaults(norm_pix_loss=False)

    # Optimizer parameters
    parser.add_argument('--weight_decay', type=float, default=0.05,
                        help='weight decay (default: 0.05)')
    parser.add_argument('--lr', type=float, default=None, metavar='LR',
                        help='learning rate (absolute lr)')
    parser.add_argument('--blr', type=float, default=1e-3, metavar='LR',
                        help='base learning rate: absolute_lr = base_lr * total_batch_size / 256')
    parser.add_argument('--min_lr', type=float, default=0., metavar='LR',
                        help='lower lr bound for cyclic schedulers that hit 0')
    parser.add_argument('--warmup_epochs', type=int, default=10, metavar='N',
                        help='epochs to warmup LR')

    # Dataset parameters
    parser.add_argument('--data_path', default='/tmp/datasets/', type=str,
                        help='dataset path')
    parser.add_argument('--anno_file', default='annotation_FSC147_384.json', type=str,
                        help='annotation json file')
    parser.add_argument('--data_split_file', default='Train_Test_Val_FSC_147.json', type=str,
                        help='data split json file')
    parser.add_argument('--class_file', default='ImageClasses_FSC147.txt', type=str,
                        help='class json file')
    parser.add_argument('--im_dir', default='images_384_VarV2', type=str,
                        help='images directory')
    parser.add_argument('--gt_dir', default='gt_density_map_adaptive_384_VarV2', type=str,
                        help='ground truth directory')
    parser.add_argument('--output_dir', default='./data/out/fim6_dir',
                        help='path where to save, empty for no saving')
    parser.add_argument('--device', default='cuda',
                        help='device to use for training / testing')
    parser.add_argument('--seed', default=0, type=int)
    parser.add_argument('--resume', default=None, type=str,
                        help='resume from checkpoint')

    # Training parameters
    parser.add_argument('--start_epoch', default=0, type=int, metavar='N',
                        help='start epoch')
    parser.add_argument('--num_workers', default=10, type=int)
    parser.add_argument('--pin_mem', action='store_true',
                        help='Pin CPU memory in DataLoader for more efficient (sometimes) transfer to GPU.')
    parser.add_argument('--no_pin_mem', action='store_false', dest='pin_mem')
    parser.set_defaults(pin_mem=True)

    # Distributed training parameters
    parser.add_argument('--world_size', default=1, type=int,
                        help='number of distributed processes')
    parser.add_argument('--local_rank', default=-1, type=int)
    parser.add_argument('--dist_on_itp', action='store_true')
    parser.add_argument('--dist_url', default='env://',
                        help='url used to set up distributed training')

    # Logging parameters
    parser.add_argument('--log_dir', default='./logs/fim6_dir',
                        help='path where to tensorboard log')
    parser.add_argument("--title", default="CounTR_finetuning", type=str)
    parser.add_argument("--wandb", default=None, type=str)
    parser.add_argument("--team", default="fdudip", type=str)
    parser.add_argument("--wandb_id", default=None, type=str)

    return parser


os.environ["CUDA_LAUNCH_BLOCKING"] = '1'


class TestData(Dataset):
    def __init__(self):

        self.img = data_split['val']
        self.img_dir = im_dir

    def __len__(self):
        return len(self.img)

    def __getitem__(self, idx):
        im_id = self.img[idx]
        anno = annotations[im_id]
        dots = np.array(anno['points'])

        image = Image.open('{}/{}'.format(im_dir, im_id))
        image.load()
        W, H = image.size

        new_H = 384
        new_W = 16 * int((W / H * 384) / 16)
        scale_factor_W = float(new_W) / W
        scale_factor_H = float(new_H) / H
        image = transforms.Resize((new_H, new_W))(image)
        Normalize = transforms.Compose([transforms.ToTensor()])
        image = Normalize(image)

        # Only for visualisation purpose, no need for ground truth density map indeed.
        gt_map = np.zeros((image.shape[1], image.shape[2]), dtype='float32')
        for i in range(dots.shape[0]):
            gt_map[min(new_H - 1, int(dots[i][1] * scale_factor_H))][min(new_W - 1, int(dots[i][0] * scale_factor_W))] = 1
        gt_map = ndimage.gaussian_filter(gt_map, sigma=(1, 1), order=0)
        gt_map = torch.from_numpy(gt_map)
        gt_map = gt_map * 60

        sample = {'image': image, 'dots': dots, 'gt_map': gt_map, 'name': im_id}
        return sample['image'], sample['dots'], sample['gt_map'], sample['name']


class TrainData(Dataset):
    def __init__(self):
        self.img = data_split['train']
        random.shuffle(self.img)
        self.img_dir = im_dir
        self.TransformTrain = transform_train(data_path)

    def __len__(self):
        return len(self.img)

    def __getitem__(self, idx):
        im_id = self.img[idx]
        anno = annotations[im_id]

        dots = np.array(anno['points'])

        image = Image.open('{}/{}'.format(im_dir, im_id))
        image.load()
        density_path = gt_dir / (im_id.split(".jpg")[0] + ".npy")
        density = np.load(density_path).astype('float32')

        sample = {'image': image, 'gt_density': density, 'dots': dots, 'id': im_id}
        sample = self.TransformTrain(sample)
        return sample['image'], sample['gt_density']


def main(args):
    misc.init_distributed_mode(args)

    print('job dir: {}'.format(os.path.dirname(os.path.realpath(__file__))))
    print("{}".format(args).replace(', ', ',\n'))

    device = torch.device(args.device)

    # fix the seed for reproducibility
    seed = args.seed + misc.get_rank()
    torch.manual_seed(seed)
    np.random.seed(seed)

    cudnn.benchmark = True

    dataset_train = TrainData()
    dataset_val_zs = TestData()

    if True:  # args.distributed:
        num_tasks = misc.get_world_size()
        global_rank = misc.get_rank()
        sampler_train = torch.utils.data.DistributedSampler(
            dataset_train, num_replicas=num_tasks, rank=global_rank, shuffle=True
        )
        sampler_val_zs = torch.utils.data.DistributedSampler(
            dataset_val_zs, num_replicas=num_tasks, rank=global_rank, shuffle=False
        )
    else:
        sampler_train = torch.utils.data.RandomSampler(dataset_train)

    loss_func = nn.MSELoss()

    log_writer = None
    if global_rank == 0:
        if args.wandb is not None:
            wandb_run = wandb.init(
                config=args,
                resume="allow",
                project=args.wandb,
                name=args.title,
                entity=args.team,
                tags=["CounTR", "finetuning"],
                id=args.wandb_id,
            )
        else:
            wandb_run = None

    data_loader_train = torch.utils.data.DataLoader(
        dataset_train, sampler=sampler_train,
        batch_size=args.batch_size,
        num_workers=args.num_workers,
        pin_memory=args.pin_mem,
        drop_last=False,
    )

    data_loader_val_zs = torch.utils.data.DataLoader(
        dataset_val_zs, sampler=sampler_val_zs,
        batch_size=1,
        num_workers=args.num_workers,
        pin_memory=args.pin_mem,
        drop_last=False,
    )

    # define the model
    model = models_mae_cross.__dict__[args.model](
        norm_pix_loss=args.norm_pix_loss)

    model.to(device)

    model_without_ddp = model

    print("Model = %s" % str(model_without_ddp))

    eff_batch_size = args.batch_size * args.accum_iter * misc.get_world_size()

    if args.lr is None:  # only base_lr is specified
        args.lr = args.blr * eff_batch_size / 256

    print("base lr: %.2e" % (args.lr * 256 / eff_batch_size))
    print("actual lr: %.2e" % args.lr)

    print("accumulate grad iterations: %d" % args.accum_iter)
    print("effective batch size: %d" % eff_batch_size)

    if args.distributed:
        model = torch.nn.parallel.DistributedDataParallel(
            model, device_ids=[args.gpu], find_unused_parameters=True)
        model_without_ddp = model.module

    # following timm: set wd as 0 for bias and norm layers
    param_groups = optim_factory.add_weight_decay(
        model_without_ddp, args.weight_decay)
    optimizer = torch.optim.AdamW(param_groups, lr=args.lr, betas=(0.9, 0.95))
    print(optimizer)

    loss_scaler = NativeScaler()
    zs_min_MAE = 99999
    misc.load_model_FSC(args=args, model_without_ddp=model_without_ddp)

    print(f"Start training for {args.epochs} epochs")
    start_time = time.time()
    for epoch in range(args.start_epoch, args.epochs):
        if args.distributed:
            data_loader_train.sampler.set_epoch(epoch)

        # train one epoch
        model.train(True)
        metric_logger = misc.MetricLogger(delimiter="  ")
        metric_logger.add_meter('lr', misc.SmoothedValue(
            window_size=1, fmt='{value:.6f}'))
        header = 'Epoch: [{}]'.format(epoch)
        print_freq = 20
        accum_iter = args.accum_iter

        # some parameters in training
        train_loss = 0
        train_rmse = 0
        pred_cnt = 0
        gt_cnt = 0

        optimizer.zero_grad()

        # if misc.is_main_process():
        #     if log_writer is not None:
        #         print('log_dir: {}'.format(log_writer.log_dir))

        for data_iter_step, (samples, gt_density) in enumerate(
                metric_logger.log_every(data_loader_train, print_freq, header)):
            epoch_1000x = int(
                (data_iter_step / len(data_loader_train) + epoch) * 1000)

            if data_iter_step % accum_iter == 0:
                lr_sched.adjust_learning_rate(
                    optimizer, data_iter_step / len(data_loader_train) + epoch, args)

            samples = samples.to(device, non_blocking=True).half()
            gt_density = gt_density.to(device, non_blocking=True).half()

            with torch.cuda.amp.autocast():
                output = model(samples)

            # Compute loss function
            loss = loss_func(output, gt_density)

            loss_value = loss.item()

            # print(f'{data_iter_step}/{len(data_loader_train)}: loss: {loss_value}')

            train_loss += loss_value

            # Output visualisation information to tensorboard
            if data_iter_step == 0 and misc.is_main_process():
                if log_writer is not None:
                    fig = output[0].unsqueeze(0).repeat(3, 1, 1)
                    f1 = gt_density[0].unsqueeze(0).repeat(3, 1, 1)

                    log_writer.add_images(
                        'gt_density', (samples[0] / 2 + f1 / 10), int(epoch), dataformats='CHW')
                    log_writer.add_images(
                        'density map', (fig / 20), int(epoch), dataformats='CHW')
                    log_writer.add_images(
                        'density map overlay', (samples[0] / 2 + fig / 10), int(epoch), dataformats='CHW')

                if wandb_run is not None:
                    wandb_densities = []

                    for i in range(samples.shape[0]):
                        fig = output[i].unsqueeze(0).repeat(3, 1, 1)
                        f1 = gt_density[i].unsqueeze(0).repeat(3, 1, 1)
                        w_densities = torch.cat(
                            [samples[i], f1, fig], dim=2)
                        w_densities = misc.min_max(w_densities)
                        wandb_densities += [wandb.Image(
                            torchvision.transforms.ToPILImage()(w_densities))]
                    wandb.log({f"Train/Density Predictions": wandb_densities},
                              step=epoch_1000x, commit=False)

            if not math.isfinite(loss_value):
                print("Loss is {}, stopping training".format(loss_value))
                sys.exit(1)

            loss /= accum_iter
            loss_scaler(loss, optimizer, parameters=model.parameters(),
                        update_grad=(data_iter_step + 1) % accum_iter == 0)
            if (data_iter_step + 1) % accum_iter == 0:
                optimizer.zero_grad()

            torch.cuda.synchronize()

            metric_logger.update(loss=loss_value)

            lr = optimizer.param_groups[0]["lr"]
            metric_logger.update(lr=lr)

            loss_value_reduce = misc.all_reduce_mean(loss_value)
            if (data_iter_step + 1) % accum_iter == 0 and misc.is_main_process():
                if log_writer is not None:
                    """ We use epoch_1000x as the x-axis in tensorboard.
                    This calibrates different curves when batch size changes.
                    """
                    log_writer.add_scalar(
                        'train_loss', loss_value_reduce, epoch_1000x)
                    log_writer.add_scalar('lr', lr, epoch_1000x)
                if wandb_run is not None:
                    log = {"Train/Loss": loss_value_reduce, "Train/LR": lr}
                    wandb.log(log, step=epoch_1000x,
                              commit=False)

        # Only use 1 batches when overfitting
        metric_logger.synchronize_between_processes()
        print("Averaged stats:", metric_logger)
        train_stats = {k: meter.global_avg for k,
                       meter in metric_logger.meters.items()}
        
        # Begin Validation
        val_zs_mae = 0
        val_zs_metric_logger = misc.MetricLogger(delimiter="  ")
        for data_iter_step, (samples, gt_dots, gt_map, im_name) in \
                enumerate(val_zs_metric_logger.log_every(data_loader_val_zs, print_freq, header)):
            im_name = Path(im_name[0])
            gt_map = gt_map.to(device, non_blocking=True)
            samples = samples.to(device, non_blocking=True)
            gt_dots = gt_dots.to(device, non_blocking=True).half()
            _, _, h, w = samples.shape
            density_map = torch.zeros([h, w])
            density_map = density_map.to(device, non_blocking=True)
            start = 0
            prev = -1
            with torch.no_grad():
                while start + 383 < w:
                    output, = model(samples[:, :, :, start:start + 384])
                    output = output.squeeze(0)
                    b1 = nn.ZeroPad2d(padding=(start, w - prev - 1, 0, 0))
                    d1 = b1(output[:, 0:prev - start + 1])
                    b2 = nn.ZeroPad2d(padding=(prev + 1, w - start - 384, 0, 0))
                    d2 = b2(output[:, prev - start + 1:384])

                    b3 = nn.ZeroPad2d(padding=(0, w - start, 0, 0))
                    density_map_l = b3(density_map[:, 0:start])
                    density_map_m = b1(density_map[:, start:prev + 1])
                    b4 = nn.ZeroPad2d(padding=(prev + 1, 0, 0, 0))
                    density_map_r = b4(density_map[:, prev + 1:w])

                    density_map = density_map_l + density_map_r + density_map_m / 2 + d1 / 2 + d2

                    prev = start + 383
                    start = start + 128
                    if start + 383 >= w:
                        if start == w - 384 + 128:
                            break
                        else:
                            start = w - 384
            cnt_err = torch.norm((density_map - gt_map), p=2)
            val_zs_mae += cnt_err
        val_zs_metric_logger.synchronize_between_processes()


        if misc.is_main_process():
            if wandb_run is not None:
                log = {"Val/Loss": val_zs_mae}
                fig = density_map.unsqueeze(0).repeat(3, 1, 1)
                f1 = gt_map[0].unsqueeze(0).repeat(3, 1, 1)
                w_densities = torch.cat(
                    [samples[0], fig, f1], dim=2)
                w_densities = misc.min_max(w_densities)
                wandb_densities = [wandb.Image(
                    torchvision.transforms.ToPILImage()(w_densities))]
                wandb.log({f"Val/Density Predictions": wandb_densities},
                            step=epoch_1000x, commit=False)
                wandb.log(log, step=epoch_1000x,
                            commit=True if data_iter_step == 0 else False)
                print(log)

        # save train status and model
        if args.output_dir and (epoch % 50 == 0 or epoch + 1 == args.epochs):
            misc.save_model(
                args=args, model=model, model_without_ddp=model_without_ddp, optimizer=optimizer,
                loss_scaler=loss_scaler, epoch=epoch, suffix=f"_{epoch}")
            
        if args.output_dir and val_zs_mae / (len(data_loader_val_zs)) < zs_min_MAE:
            zs_min_MAE = val_zs_mae / (len(data_loader_val_zs))
            misc.save_model(
                args=args, model=model, model_without_ddp=model_without_ddp, optimizer=optimizer,
                loss_scaler=loss_scaler, epoch=epoch, suffix="_zs_min")

        # Output log status
        log_stats = {**{f'train_{k}': v for k, v in train_stats.items()},
                     'Current MAE': train_loss / (len(data_loader_train) * args.batch_size),
                     'RMSE': (train_rmse / (len(data_loader_train) * args.batch_size)) ** 0.5,
                     'epoch': epoch, }

        print('Current MAE: {:5.2f}, RMSE: {:5.2f} '.format(train_loss / (len(data_loader_train) * args.batch_size), (
            train_rmse / (len(data_loader_train) * args.batch_size)) ** 0.5))

        if args.output_dir and misc.is_main_process():
            if log_writer is not None:
                log_writer.flush()
            with open(os.path.join(args.output_dir, "log.txt"), mode="a", encoding="utf-8") as f:
                f.write(json.dumps(log_stats) + "\n")

    total_time = time.time() - start_time
    total_time_str = str(datetime.timedelta(seconds=int(total_time)))
    print('Training time {}'.format(total_time_str))
    if misc.is_main_process():
        wandb.run.finish()


if __name__ == '__main__':
    args = get_args_parser()
    args = args.parse_args()

    # load data from FSC147
    data_path = Path(args.data_path)
    anno_file = data_path / args.anno_file
    data_split_file = data_path / args.data_split_file
    im_dir = data_path / args.im_dir
    gt_dir = data_path / args.gt_dir
    class_file = data_path / args.class_file
    with open(anno_file) as f:
        annotations = json.load(f)
    with open(data_split_file) as f:
        data_split = json.load(f)
    class_dict = {}
    with open(class_file) as f:
        for line in f:
            key = line.split()[0]
            val = line.split()[1:]
            class_dict[key] = val

    if args.output_dir:
        Path(args.output_dir).mkdir(parents=True, exist_ok=True)
    main(args)