validate_ood.py

#!/usr/bin/env python3
# Copyright (c) 2021-2022, NVIDIA Corporation & Affiliates. All rights reserved.
#
# This work is made available under the Nvidia Source Code License-NC.
# To view a copy of this license, visit
# https://github.com/NVlabs/FAN/blob/main/LICENSE

""" ImageNet Validation Script
This is intended to be a lean and easily modifiable ImageNet validation script for evaluating pretrained
models or training checkpoints against ImageNet or similarly organized image datasets. It prioritizes
canonical PyTorch, standard Python style, and good performance. Repurpose as you see fit.
Hacked together by Ross Wightman (https://github.com/rwightman)
"""

import argparse
import errno
import os
import csv
import glob
import time
import logging
import torch
import torch.nn as nn
import torch.nn.parallel
from collections import OrderedDict
from contextlib import suppress

from timm.models import create_model, apply_test_time_pool, resume_checkpoint, load_checkpoint, is_model, list_models
from timm.data import resolve_data_config, RealLabelsImagenet
from timm.utils import accuracy, AverageMeter, natural_key, setup_default_logging, set_jit_legacy

import numpy as np



from utils.imagenet_a import indices_in_1k
from utils.imagenet_r import imagenet_r_mask
from utils.mce_utils import get_ce_alexnet, get_mce_from_accuracy


from data import create_loader, create_dataset
from optim_factory import create_optimizer_v2, optimizer_kwargs
from models import vision_transformer, swin_transformer, convnext



has_apex = False
try:
    from apex import amp
    has_apex = True
except ImportError:
    pass

has_native_amp = False
try:
    if getattr(torch.cuda.amp, 'autocast') is not None:
        has_native_amp = True
except AttributeError:
    pass

torch.backends.cudnn.benchmark = True
_logger = logging.getLogger('validate')


parser = argparse.ArgumentParser(description='PyTorch ImageNet Validation')
parser.add_argument('data', metavar='DIR',
                    help='path to dataset')
parser.add_argument('--dataset', '-d', metavar='NAME', default='',
                    help='dataset type (default: ImageFolder/ImageTar if empty)')
parser.add_argument('--split', metavar='NAME', default='validation',
                    help='dataset split (default: validation)')
parser.add_argument('--model', '-m', metavar='NAME', default='dpn92',
                    help='model architecture (default: dpn92)')
parser.add_argument('-j', '--workers', default=4, type=int, metavar='N',
                    help='number of data loading workers (default: 2)')
parser.add_argument('-b', '--batch-size', default=256, type=int,
                    metavar='N', help='mini-batch size (default: 256)')
parser.add_argument('--img-size', default=None, type=int,
                    metavar='N', help='Input image dimension, uses model default if empty')
parser.add_argument('--input-size', default=None, nargs=3, type=int,
                    metavar='N N N', help='Input all image dimensions (d h w, e.g. --input-size 3 224 224), uses model default if empty')
parser.add_argument('--crop-pct', default=None, type=float,
                    metavar='N', help='Input image center crop pct')
parser.add_argument('--mean', type=float, nargs='+', default=None, metavar='MEAN',
                    help='Override mean pixel value of dataset')
parser.add_argument('--std', type=float,  nargs='+', default=None, metavar='STD',
                    help='Override std deviation of of dataset')
parser.add_argument('--interpolation', default='', type=str, metavar='NAME',
                    help='Image resize interpolation type (overrides model)')
parser.add_argument('--num-classes', type=int, default=None,
                    help='Number classes in dataset')
parser.add_argument('--class-map', default='', type=str, metavar='FILENAME',
                    help='path to class to idx mapping file (default: "")')
parser.add_argument('--gp', default=None, type=str, metavar='POOL',
                    help='Global pool type, one of (fast, avg, max, avgmax, avgmaxc). Model default if None.')
parser.add_argument('--log-freq', default=50, type=int,
                    metavar='N', help='batch logging frequency (default: 10)')
parser.add_argument('--checkpoint', default='', type=str, metavar='PATH',
                    help='path to latest checkpoint (default: none)')
parser.add_argument('--pretrained', dest='pretrained', action='store_true',
                    help='use pre-trained model')
parser.add_argument('--num-gpu', type=int, default=1,
                    help='Number of GPUS to use')
parser.add_argument('--no-test-pool', dest='no_test_pool', action='store_true',
                    help='disable test time pool')
parser.add_argument('--no-prefetcher', action='store_true', default=False,
                    help='disable fast prefetcher')
parser.add_argument('--pin-mem', action='store_true', default=False,
                    help='Pin CPU memory in DataLoader for more efficient (sometimes) transfer to GPU.')
parser.add_argument('--channels-last', action='store_true', default=False,
                    help='Use channels_last memory layout')
parser.add_argument('--amp', action='store_true', default=False,
                    help='Use AMP mixed precision. Defaults to Apex, fallback to native Torch AMP.')
parser.add_argument('--apex-amp', action='store_true', default=False,
                    help='Use NVIDIA Apex AMP mixed precision')
parser.add_argument('--native-amp', action='store_true', default=False,
                    help='Use Native Torch AMP mixed precision')
parser.add_argument('--tf-preprocessing', action='store_true', default=False,
                    help='Use Tensorflow preprocessing pipeline (require CPU TF installed')
parser.add_argument('--use-ema', dest='use_ema', action='store_true',
                    help='use ema version of weights if present')
parser.add_argument('--torchscript', dest='torchscript', action='store_true',
                    help='convert model torchscript for inference')
parser.add_argument('--legacy-jit', dest='legacy_jit', action='store_true',
                    help='use legacy jit mode for pytorch 1.5/1.5.1/1.6 to get back fusion performance')
parser.add_argument('--results-file', default='', type=str, metavar='FILENAME',
                    help='Output csv file for validation results (summary)')
parser.add_argument('--real-labels', default='', type=str, metavar='FILENAME',
                    help='Real labels JSON file for imagenet evaluation')
parser.add_argument('--valid-labels', default='', type=str, metavar='FILENAME',
                    help='Valid label indices txt file for validation of partial label space')

# finetuning
parser.add_argument('--tuning-mode', default=None, type=str,
                    help='Method of fine-tuning (default: None')
parser.add_argument('--num-vpt', default=None, type=int,
                    help='The number of prompts in VPT')


parser.add_argument('--imagenet_a', action='store_true', default=False,
                    help='replace labels from 1k to 200')
parser.add_argument('--imagenet_r', action='store_true', default=False,
                    help='replace labels from 1k to imagenet-r indices')
parser.add_argument('--imagenet_c', action='store_true', default=False,
                    help='use corrupted dataset for evaluation')

def validate(args):
    args.pretrained = args.pretrained or not args.checkpoint
    args.prefetcher = not args.no_prefetcher
    amp_autocast = suppress  # do nothing
    if args.amp:
        if has_native_amp:
            args.native_amp = True
        elif has_apex:
            args.apex_amp = True
        else:
            _logger.warning("Neither APEX or Native Torch AMP is available.")
    assert not args.apex_amp or not args.native_amp, "Only one AMP mode should be set."
    if args.native_amp:
        amp_autocast = torch.cuda.amp.autocast
        _logger.info('Validating in mixed precision with native PyTorch AMP.')
    elif args.apex_amp:
        _logger.info('Validating in mixed precision with NVIDIA APEX AMP.')
    else:
        _logger.info('Validating in float32. AMP not enabled.')

    if args.legacy_jit:
        set_jit_legacy()


    model = create_model(
        args.model,
        pretrained=args.pretrained,
        num_classes=args.num_classes,
        global_pool=args.gp,
        scriptable=args.torchscript,
        tuning_mode=args.tuning_mode,
        num_vpt=args.num_vpt
        )


    if args.num_classes is None:
        assert hasattr(model, 'num_classes'), 'Model must have `num_classes` attr if not set on cmd line/config.'
        args.num_classes = model.num_classes

    if args.checkpoint:
        resume_epoch = resume_checkpoint(
            model, args.checkpoint
            )

    param_count = sum([m.numel() for m in model.parameters()])
    _logger.info('Model %s created, param count: %d' % (args.model, param_count))

    data_config = resolve_data_config(vars(args), model=model, use_test_size=True)
    test_time_pool = False
    if not args.no_test_pool:
        model, test_time_pool = apply_test_time_pool(model, data_config, use_test_size=True)

    if args.torchscript:
        torch.jit.optimized_execution(True)
        model = torch.jit.script(model)

    model = model.cuda()
    if args.apex_amp:
        model = amp.initialize(model, opt_level='O1')

    if args.channels_last:
        model = model.to(memory_format=torch.channels_last)

    if args.num_gpu > 1:
        model = torch.nn.DataParallel(model, device_ids=list(range(args.num_gpu)))

    criterion = nn.CrossEntropyLoss().cuda()

    print(args.data)
    dataset = create_dataset(
        root=args.data, name=args.dataset, split=args.split,
        load_bytes=args.tf_preprocessing, class_map=args.class_map)

    if args.valid_labels:
        with open(args.valid_labels, 'r') as f:
            valid_labels = {int(line.rstrip()) for line in f}
            valid_labels = [i in valid_labels for i in range(args.num_classes)]
    else:
        valid_labels = None

    if args.real_labels:
        real_labels = RealLabelsImagenet(dataset.filenames(basename=True), real_json=args.real_labels)
    else:
        real_labels = None

    crop_pct = 1.0 if test_time_pool else data_config['crop_pct']
    loader = create_loader(
        dataset,
        input_size=data_config['input_size'],
        batch_size=args.batch_size,
        use_prefetcher=args.prefetcher,
        interpolation=data_config['interpolation'],
        mean=data_config['mean'],
        std=data_config['std'],
        num_workers=args.workers,
        crop_pct=crop_pct,
        pin_memory=args.pin_mem,
        tf_preprocessing=args.tf_preprocessing)

    batch_time = AverageMeter()
    losses = AverageMeter()
    top1 = AverageMeter()
    top5 = AverageMeter()

    model.eval()
    with torch.no_grad():
        input = torch.randn((args.batch_size,) + data_config['input_size']).cuda()
        if args.channels_last:
            input = input.contiguous(memory_format=torch.channels_last)
        end = time.time()
        for batch_idx, (input, target) in enumerate(loader):
            if args.no_prefetcher:
                target = target.cuda()
                input = input.cuda()
            if args.channels_last:
                input = input.contiguous(memory_format=torch.channels_last)

            # compute output
            with amp_autocast():
                output = model(input)
                if args.imagenet_a:
                    output = output[:, indices_in_1k]
                if args.imagenet_r:
                    output = output[:, imagenet_r_mask]
            if isinstance(output, (tuple, list)):
                output = output[0]
            if valid_labels is not None:
                output = output[:, valid_labels]
            loss = criterion(output, target)

            if real_labels is not None:
                real_labels.add_result(output)

            # measure accuracy and record loss
            acc1, acc5 = accuracy(output.detach(), target, topk=(1, 5))
            losses.update(loss.item(), input.size(0))
            top1.update(acc1.item(), input.size(0))
            top5.update(acc5.item(), input.size(0))

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

            if batch_idx % args.log_freq == 0:
                _logger.info(
                    'Test: [{0:>4d}/{1}]  '
                    'Time: {batch_time.val:.3f}s ({batch_time.avg:.3f}s, {rate_avg:>7.2f}/s)  '
                    'Loss: {loss.val:>7.4f} ({loss.avg:>6.4f})  '
                    'Acc@1: {top1.val:>7.3f} ({top1.avg:>7.3f})  '
                    'Acc@5: {top5.val:>7.3f} ({top5.avg:>7.3f})'.format(
                        batch_idx, len(loader), batch_time=batch_time,
                        rate_avg=input.size(0) / batch_time.avg,
                        loss=losses, top1=top1, top5=top5))

    if real_labels is not None:
        # real labels mode replaces topk values at the end
        top1a, top5a = real_labels.get_accuracy(k=1), real_labels.get_accuracy(k=5)
    else:
        top1a, top5a = top1.avg, top5.avg
    results = OrderedDict(
        top1=round(top1a, 4), top1_err=round(100 - top1a, 4),
        top5=round(top5a, 4), top5_err=round(100 - top5a, 4),
        param_count=round(param_count / 1e6, 2),
        img_size=data_config['input_size'][-1],
        cropt_pct=crop_pct,
        interpolation=data_config['interpolation'])

    _logger.info(' * Acc@1 {:.3f} ({:.3f}) Acc@5 {:.3f} ({:.3f})'.format(
       results['top1'], results['top1_err'], results['top5'], results['top5_err']))

    return results


def main():
    setup_default_logging()
    args = parser.parse_args()
    if not args.imagenet_c:
        if args.imagenet_a or args.imagenet_r:
            validate(args)
        else:
            print('Please specify an OOD dataset.')
            return
    else:
        results_file = args.results_file or './results-all.csv'
        os.makedirs(results_file, exist_ok=True)
        blur_list = ['gaussian_blur', 'motion_blur', 'glass_blur', 'defocus_blur']
        noise_list = ['gaussian_noise', 'shot_noise', 'speckle_noise', 'impulse_noise']
        digital_list = ['contrast', 'jpeg_compression', 'saturate', 'pixelate']
        weather_list = ['snow', 'fog', 'frost', 'spatter', 'brightness']
        extra = ['zoom_blur', 'elastic_transform']
        name_list = noise_list + extra + blur_list + digital_list + weather_list
        ce_alexnet = get_ce_alexnet()
        mCE = 0
        counter = 0
        average_acc = {}
        base_dir = args.data
        for noise_name in name_list:
            res_sum = 0
            root = base_dir + noise_name + '/'
            results = []
            for i in range(0, 5):
                args.data = root + str(i+1)
                print('validating dir:', args.data)
                res = validate(args)
                results.append(res['top1'])
                res_sum += res['top1']
                if noise_name in ce_alexnet.keys():
                    CE = get_mce_from_accuracy(res['top1'], ce_alexnet[noise_name])
                    mCE += CE
                    counter += 1
            results.append(res_sum/(i+1))
            average_acc[noise_name] = res_sum/(i+1)
            np.savetxt(results_file  + noise_name + '_' + '%.2f' % (res_sum/(i+1)) + '.csv', results)
            print('average score is:', res_sum / (i+1))
            print('current mCE is: ', mCE/counter)
        np.savetxt(results_file + 'mCE' + '_' + '%.2f' % (mCE/counter) + '.csv', results)
        print('all average score is:', average_acc)
        print('mCE is: ', mCE/counter)


def write_results(results_file, results):
    with open(results_file, mode='w') as cf:
        dw = csv.DictWriter(cf, fieldnames=results[0].keys())
        dw.writeheader()
        for r in results:
            dw.writerow(r)
        cf.flush()

if __name__ == '__main__':
    main()