utils.py

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

import os
import torch
import torch.distributed as dist
from torch._six import inf
import argparse
from config import get_config
import torch.nn.functional as F
import numpy as np

def parse_option():
    parser = argparse.ArgumentParser('Swin Transformer training and evaluation script', add_help=False)
    parser.add_argument('--cfg', type=str, required=True, metavar="FILE", help='path to config file', )
    parser.add_argument(
        "--opts",
        help="Modify config options by adding 'KEY VALUE' pairs. ",
        default=None,
        nargs='+',
    )

    # easy config modification
    parser.add_argument('--batch-size', type=int, help="batch size for single GPU")
    parser.add_argument('--eval-batch-size', type=int, help="batch size for single GPU")
    parser.add_argument('--data-path', type=str, help='path to dataset')
    parser.add_argument('--zip', action='store_true', help='use zipped dataset instead of folder dataset')
    parser.add_argument('--cache-mode', type=str, default='part', choices=['no', 'full', 'part'],
                        help='no: no cache, '
                             'full: cache all data, '
                             'part: sharding the dataset into nonoverlapping pieces and only cache one piece')
    parser.add_argument('--pretrained',
                        help='pretrained weight from checkpoint, could be imagenet22k pretrained weight')
    parser.add_argument('--resume', help='resume from checkpoint')
    parser.add_argument('--accumulation-steps', type=int, help="gradient accumulation steps")
    parser.add_argument('--use-checkpoint', action='store_true',
                        help="whether to use gradient checkpointing to save memory")
    parser.add_argument('--disable_amp', action='store_true', help='Disable pytorch amp')
    parser.add_argument('--amp-opt-level', type=str, choices=['O0', 'O1', 'O2'],
                        help='mixed precision opt level, if O0, no amp is used (deprecated!)')
    parser.add_argument('--output', default='output', type=str, metavar='PATH',
                        help='root of output folder, the full path is <output>/<model_name>/<tag> (default: output)')
    parser.add_argument('--tag', help='tag of experiment')
    parser.add_argument('--eval', action='store_true', help='Perform evaluation only')
    parser.add_argument('--throughput', action='store_true', help='Test throughput only')

    # distributed training
    parser.add_argument("--local_rank", type=int, required=True, help='local rank for DistributedDataParallel')

    # for acceleration
    parser.add_argument('--fused_window_process', action='store_true',
                        help='Fused window shift & window partition, similar for reversed part.')
    parser.add_argument('--fused_layernorm', action='store_true', help='Use fused layernorm.')
    ## overwrite optimizer in config (*.yaml) if specified, e.g., fused_adam/fused_lamb
    parser.add_argument('--optim', type=str,
                        help='overwrite optimizer if provided, can be adamw/sgd/fused_adam/fused_lamb.')
    parser.add_argument('--model', type=str,
                        help='method used, can be base, swap_only,mask_only or full.')
    parser.add_argument('--swap_weight', type=float, default=0.5, help='swap weight')
    parser.add_argument('--mask', action='store_true',
                        help="whether to mask image")
    parser.add_argument('--swap_w', type=float, default=0.5,
                        help="mask weight for loss")
    parser.add_argument('--con_w', type=float, default=0.5,
                        help="loss weight for con loss")
    parser.add_argument('--origin_w', type=float, default=1,
                        help="weight for original loss")
    parser.add_argument('--swap', action='store_true',
                        help="whether to swap")
    parser.add_argument("--num_part", type=int, default=2,
                        help="local_rank for distributed training on gpus")
    parser.add_argument("--img_size", type=int, default=224,
                        help="local_rank for distributed training on gpus")
    parser.add_argument('--mix', action='store_true',
                        help="whether to mix")

    parser.add_argument('--con', action='store_true',
                        help="whether to apply contrastive loss")

    parser.add_argument("--epochs", type=int, default=200,
                        help="Training epochs")

    parser.add_argument('--lr', type=float, default=5e-4,
                        help="base learning rate")
    parser.add_argument('--head_mul', type=float, default=5.0,
                        help="base learning rate head mul")

    parser.add_argument('--margin', type=float, default=0.4,
                        help="base learning rate")

    parser.add_argument('--dataset', type=str, help="dataset used.")

    args, unparsed = parser.parse_known_args()

    config = get_config(args)

    return args, config


def my_con_loss(features, labels, margin=0.4):
    B, _ = features.shape
    features = F.normalize(features)
    cos_matrix = features.mm(features.t())
    pos_label_matrix = torch.stack([labels == labels[i] for i in range(B)]).float()
    neg_label_matrix = 1 - pos_label_matrix
    pos_cos_matrix = 1 - cos_matrix
    neg_cos_matrix = cos_matrix - margin
    neg_cos_matrix[neg_cos_matrix < 0] = 0
    loss = (pos_cos_matrix * pos_label_matrix).sum() + (neg_cos_matrix * neg_label_matrix).sum()
    loss /= (B * B)
    return loss


def con_loss(features, labels, margin=0.4):
    B, _ = features.shape
    features = F.normalize(features)
    cos_matrix = features.mm(features.t())
    pos_label_matrix = torch.stack([labels == labels[i] for i in range(B)]).float()
    neg_label_matrix = 1 - pos_label_matrix
    pos_cos_matrix = 1 - cos_matrix
    neg_cos_matrix = cos_matrix - margin
    neg_cos_matrix[neg_cos_matrix < 0] = 0
    loss = (pos_cos_matrix * pos_label_matrix).sum() + (neg_cos_matrix * neg_label_matrix).sum()
    loss /= (B * B)
    return loss


def instance_con_loss(features, labels, margin=1.0):
    B, _ = features.shape
    half = B//2
    index = np.arange(half)
    permute_idx = np.random.permutation(B)[:half]
    while (index == permute_idx).sum() != 0 or (index == permute_idx//2).sum() != 0:
        permute_idx = np.random.permutation(B)[:half]
    mask_feats, shuffle_feats = features[:half], features[half:]
    neg_feats = features[permute_idx]
    cos_dis = F.normalize(mask_feats - shuffle_feats)
    neg_dis = F.normalize(mask_feats - neg_feats)
    # cos_matrix_pos = F.normalize(mask_feats - shuffle_feats)
    # cos_matrix_neg = F.normalize(mask_feats - neg_feats)
    loss = cos_dis.sum(dim=-1) - neg_dis.sum(dim=-1) + margin
    # pos_label_matrix = torch.stack([labels == labels[i] for i in range(B)]).float()
    # neg_label_matrix = 1 - pos_label_matrix
    # pos_cos_matrix = 1 - cos_matrix
    # neg_cos_matrix = cos_matrix - margin
    # neg_cos_matrix[neg_cos_matrix < 0] = 0
    loss[loss < 0] = 0
    loss = loss.sum()/B
    return loss

def load_checkpoint(config, model, optimizer, lr_scheduler, loss_scaler, logger):
    logger.info(f"==============> Resuming form {config.MODEL.RESUME}....................")
    if config.MODEL.RESUME.startswith('https'):
        checkpoint = torch.hub.load_state_dict_from_url(
            config.MODEL.RESUME, map_location='cpu', check_hash=True)
    else:
        checkpoint = torch.load(config.MODEL.RESUME, map_location='cpu')
    msg = model.load_state_dict(checkpoint['model'], strict=False)
    logger.info(msg)
    max_accuracy = 0.0
    if not config.EVAL_MODE and 'optimizer' in checkpoint and 'lr_scheduler' in checkpoint and 'epoch' in checkpoint:
        optimizer.load_state_dict(checkpoint['optimizer'])
        lr_scheduler.load_state_dict(checkpoint['lr_scheduler'])
        config.defrost()
        config.TRAIN.START_EPOCH = checkpoint['epoch'] + 1
        config.freeze()
        if 'scaler' in checkpoint:
            loss_scaler.load_state_dict(checkpoint['scaler'])
        logger.info(f"=> loaded successfully '{config.MODEL.RESUME}' (epoch {checkpoint['epoch']})")
        if 'max_accuracy' in checkpoint:
            max_accuracy = checkpoint['max_accuracy']

    del checkpoint
    torch.cuda.empty_cache()
    return max_accuracy


def load_pretrained(config, model, logger):
    logger.info(f"==============> Loading weight {config.MODEL.PRETRAINED} for fine-tuning......")
    checkpoint = torch.load(config.MODEL.PRETRAINED, map_location='cpu')
    state_dict = checkpoint['model']

    # delete relative_position_index since we always re-init it
    relative_position_index_keys = [k for k in state_dict.keys() if "relative_position_index" in k]
    for k in relative_position_index_keys:
        del state_dict[k]

    # delete relative_coords_table since we always re-init it
    relative_position_index_keys = [k for k in state_dict.keys() if "relative_coords_table" in k]
    for k in relative_position_index_keys:
        del state_dict[k]

    # delete attn_mask since we always re-init it
    attn_mask_keys = [k for k in state_dict.keys() if "attn_mask" in k]
    for k in attn_mask_keys:
        del state_dict[k]

    # bicubic interpolate relative_position_bias_table if not match
    relative_position_bias_table_keys = [k for k in state_dict.keys() if "relative_position_bias_table" in k]
    for k in relative_position_bias_table_keys:
        relative_position_bias_table_pretrained = state_dict[k]
        relative_position_bias_table_current = model.state_dict()[k]
        L1, nH1 = relative_position_bias_table_pretrained.size()
        L2, nH2 = relative_position_bias_table_current.size()
        if nH1 != nH2:
            logger.warning(f"Error in loading {k}, passing......")
        else:
            if L1 != L2:
                # bicubic interpolate relative_position_bias_table if not match
                S1 = int(L1 ** 0.5)
                S2 = int(L2 ** 0.5)
                relative_position_bias_table_pretrained_resized = torch.nn.functional.interpolate(
                    relative_position_bias_table_pretrained.permute(1, 0).view(1, nH1, S1, S1), size=(S2, S2),
                    mode='bicubic')
                state_dict[k] = relative_position_bias_table_pretrained_resized.view(nH2, L2).permute(1, 0)

    # bicubic interpolate absolute_pos_embed if not match
    absolute_pos_embed_keys = [k for k in state_dict.keys() if "absolute_pos_embed" in k]
    for k in absolute_pos_embed_keys:
        # dpe
        absolute_pos_embed_pretrained = state_dict[k]
        absolute_pos_embed_current = model.state_dict()[k]
        _, L1, C1 = absolute_pos_embed_pretrained.size()
        _, L2, C2 = absolute_pos_embed_current.size()
        if C1 != C1:
            logger.warning(f"Error in loading {k}, passing......")
        else:
            if L1 != L2:
                S1 = int(L1 ** 0.5)
                S2 = int(L2 ** 0.5)
                absolute_pos_embed_pretrained = absolute_pos_embed_pretrained.reshape(-1, S1, S1, C1)
                absolute_pos_embed_pretrained = absolute_pos_embed_pretrained.permute(0, 3, 1, 2)
                absolute_pos_embed_pretrained_resized = torch.nn.functional.interpolate(
                    absolute_pos_embed_pretrained, size=(S2, S2), mode='bicubic')
                absolute_pos_embed_pretrained_resized = absolute_pos_embed_pretrained_resized.permute(0, 2, 3, 1)
                absolute_pos_embed_pretrained_resized = absolute_pos_embed_pretrained_resized.flatten(1, 2)
                state_dict[k] = absolute_pos_embed_pretrained_resized

    # check classifier, if not match, then re-init classifier to zero
    head_bias_pretrained = state_dict['head.bias']
    Nc1 = head_bias_pretrained.shape[0]
    Nc2 = model.head.bias.shape[0]
    if (Nc1 != Nc2):
        if Nc1 == 21841 and Nc2 == 1000:
            logger.info("loading ImageNet-22K weight to ImageNet-1K ......")
            map22kto1k_path = f'data/map22kto1k.txt'
            with open(map22kto1k_path) as f:
                map22kto1k = f.readlines()
            map22kto1k = [int(id22k.strip()) for id22k in map22kto1k]
            state_dict['head.weight'] = state_dict['head.weight'][map22kto1k, :]
            state_dict['head.bias'] = state_dict['head.bias'][map22kto1k]
        else:
            torch.nn.init.constant_(model.head.bias, 0.)
            torch.nn.init.constant_(model.head.weight, 0.)
            del state_dict['head.weight']
            del state_dict['head.bias']
            logger.warning(f"Error in loading classifier head, re-init classifier head to 0")

    msg = model.load_state_dict(state_dict, strict=False)
    logger.warning(msg)

    logger.info(f"=> loaded successfully '{config.MODEL.PRETRAINED}'")

    del checkpoint
    torch.cuda.empty_cache()


def save_checkpoint(config, epoch, model, max_accuracy, optimizer, lr_scheduler, loss_scaler, logger):
    save_state = {'model': model.state_dict(),
                  'optimizer': optimizer.state_dict(),
                  'lr_scheduler': lr_scheduler.state_dict(),
                  'max_accuracy': max_accuracy,
                  'scaler': loss_scaler.state_dict(),
                  'epoch': epoch,
                  'config': config}

    # save_path = os.path.join(config.OUTPUT, f'ckpt_epoch_{epoch}.pth')
    save_path = os.path.join(config.OUTPUT, f'best.pth')
    logger.info(f"{save_path} saving......")
    torch.save(save_state, save_path)
    logger.info(f"{save_path} saved !!!")


def get_grad_norm(parameters, norm_type=2):
    if isinstance(parameters, torch.Tensor):
        parameters = [parameters]
    parameters = list(filter(lambda p: p.grad is not None, parameters))
    norm_type = float(norm_type)
    total_norm = 0
    for p in parameters:
        param_norm = p.grad.data.norm(norm_type)
        total_norm += param_norm.item() ** norm_type
    total_norm = total_norm ** (1. / norm_type)
    return total_norm


def auto_resume_helper(output_dir):
    checkpoints = os.listdir(output_dir)
    checkpoints = [ckpt for ckpt in checkpoints if ckpt.endswith('pth')]
    print(f"All checkpoints founded in {output_dir}: {checkpoints}")
    if len(checkpoints) > 0:
        latest_checkpoint = max([os.path.join(output_dir, d) for d in checkpoints], key=os.path.getmtime)
        print(f"The latest checkpoint founded: {latest_checkpoint}")
        resume_file = latest_checkpoint
    else:
        resume_file = None
    return resume_file


def reduce_tensor(tensor):
    rt = tensor.clone()
    dist.all_reduce(rt, op=dist.ReduceOp.SUM)
    rt /= dist.get_world_size()
    return rt


def ampscaler_get_grad_norm(parameters, norm_type: float = 2.0) -> torch.Tensor:
    if isinstance(parameters, torch.Tensor):
        parameters = [parameters]
    parameters = [p for p in parameters if p.grad is not None]
    norm_type = float(norm_type)
    if len(parameters) == 0:
        return torch.tensor(0.)
    device = parameters[0].grad.device
    if norm_type == inf:
        total_norm = max(p.grad.detach().abs().max().to(device) for p in parameters)
    else:
        total_norm = torch.norm(torch.stack([torch.norm(p.grad.detach(),
                                                        norm_type).to(device) for p in parameters]), norm_type)
    return total_norm


class NativeScalerWithGradNormCount:
    state_dict_key = "amp_scaler"

    def __init__(self):
        self._scaler = torch.cuda.amp.GradScaler()

    def __call__(self, loss, optimizer, clip_grad=None, parameters=None, create_graph=False, update_grad=True):
        self._scaler.scale(loss).backward(create_graph=create_graph)
        if update_grad:
            if clip_grad is not None:
                assert parameters is not None
                self._scaler.unscale_(optimizer)  # unscale the gradients of optimizer's assigned params in-place
                norm = torch.nn.utils.clip_grad_norm_(parameters, clip_grad)
            else:
                self._scaler.unscale_(optimizer)
                norm = ampscaler_get_grad_norm(parameters)
            self._scaler.step(optimizer)
            self._scaler.update()
        else:
            norm = None
        return norm

    def state_dict(self):
        return self._scaler.state_dict()

    def load_state_dict(self, state_dict):
        self._scaler.load_state_dict(state_dict)