test.py

import sys
from functools import partial

from typing import Callable
from typing import Dict
from typing import Tuple
from typing import Union
from argparse import Namespace

sys.path.append("vision/references/segmentation")

import presets
import torch
import torch.utils.data
import torchvision
import utils
from torch import nn
from common import flops_calculation_function
from common import NanSafeConfusionMatrix as ConfusionMatrix
from common import get_coco


def get_dataset(args: Namespace, is_train: bool, transform: Callable = None) -> Tuple[torch.utils.data.Dataset, int]:
    def sbd(*args, **kwargs):
        kwargs.pop("use_v2")
        return torchvision.datasets.SBDataset(*args, mode="segmentation", **kwargs)

    def voc(*args, **kwargs):
        kwargs.pop("use_v2")
        return torchvision.datasets.VOCSegmentation(*args, **kwargs)

    paths = {
        "voc": (args.data_path, voc, 21),
        "voc_aug": (args.data_path, sbd, 21),
        "coco": (args.data_path, get_coco, 21),
        "coco_orig": (args.data_path, partial(get_coco, use_orig=True), 81)
    }
    p, ds_fn, num_classes = paths["coco_orig"]

    if transform is None:
        transform = get_transform(is_train, args)
    image_set = "train" if is_train else "val"
    ds = ds_fn(p, image_set=image_set, transforms=transform, use_v2=args.use_v2)
    return ds, num_classes


def get_transform(is_train: bool, args: Namespace) -> Callable:
    return presets.SegmentationPresetEval(base_size=520, backend=args.backend, use_v2=args.use_v2)


def criterion(inputs: Dict[str, torch.Tensor], target: Dict[str, torch.Tensor]) -> torch.Tensor:
    losses = {}
    for name, x in inputs.items():
        losses[name] = nn.functional.cross_entropy(x, target, ignore_index=255)

    if len(losses) == 1:
        return losses["out"]

    return losses["out"] + 0.5 * losses["aux"]


def evaluate(
        model: torch.nn.Module,
        data_loader: torch.utils.data.DataLoader,
        device: Union[str, torch.device],
        num_classes: int,
        criterion: Callable,
) -> Tuple[ConfusionMatrix, float]:
    model.eval()
    confmat = ConfusionMatrix(num_classes)
    metric_logger = utils.MetricLogger(delimiter="  ")
    header = "Test:"
    num_processed_samples = 0
    with torch.inference_mode():
        for batch_n, (image, target) in enumerate(metric_logger.log_every(data_loader, 100, header)):
            image, target = image.to(device), target.to(device)
            output = model(image)
            loss = criterion(output, target)
            output = output["out"]

            confmat.update(target.flatten(), output.argmax(1).flatten())
            # FIXME need to take into account that the datasets
            # could have been padded in distributed setup
            num_processed_samples += image.shape[0]

            metric_logger.update(loss=loss.item())

        confmat.reduce_from_all_processes()

    return confmat, metric_logger.loss.global_avg


def main(args):
    if args.backend.lower() != "pil" and not args.use_v2:
        # TODO: Support tensor backend in V1?
        raise ValueError("Use --use-v2 if you want to use the tv_tensor or tensor backend.")
    if args.use_v2:
        raise ValueError("v2 is only supported for coco dataset for now.")

    print(args)

    device = torch.device(args.device)

    if args.use_deterministic_algorithms:
        torch.backends.cudnn.benchmark = False
        torch.use_deterministic_algorithms(True)
    else:
        torch.backends.cudnn.benchmark = True

    dataset_test, num_classes = get_dataset(args, is_train=False)

    test_sampler = torch.utils.data.SequentialSampler(dataset_test)

    data_loader_test = torch.utils.data.DataLoader(
        dataset_test, batch_size=1, sampler=test_sampler, num_workers=args.workers, collate_fn=utils.collate_fn
    )

    checkpoint = torch.load(args.model_path)
    model = checkpoint["model"]
    model.to(device)
    model_flops = flops_calculation_function(model=model, input_sample=next(iter(data_loader_test))[0].to(device))
    print(f"Model Flops: {model_flops}M")

    # We disable the cudnn benchmarking because it can noticeably affect the accuracy
    torch.backends.cudnn.benchmark = False
    torch.backends.cudnn.deterministic = True
    confmat, loss = evaluate(
        model=model,
        data_loader=data_loader_test,
        device=device,
        num_classes=num_classes,
        criterion=criterion,
    )
    print(confmat)
    return

def get_args_parser(add_help=True):
    import argparse

    parser = argparse.ArgumentParser(description="PyTorch Segmentation Training", add_help=add_help)

    parser.add_argument("--data-path", default="/datasets01/COCO/022719/", type=str, help="dataset path")
    parser.add_argument("--device", default="cuda", type=str, help="device (Use cuda or cpu Default: cuda)")
    parser.add_argument(
        "-b", "--batch-size", default=8, type=int, help="images per gpu, the total batch size is $NGPU x batch_size"
    )
    parser.add_argument("--epochs", default=30, type=int, metavar="N", help="number of total epochs to run")

    parser.add_argument(
        "-j", "--workers", default=16, type=int, metavar="N", help="number of data loading workers (default: 16)"
    )
    parser.add_argument(
        "--use-deterministic-algorithms", action="store_true", help="Forces the use of deterministic algorithms only."
    )
    # distributed training parameters

    parser.add_argument("--backend", default="PIL", type=str.lower, help="PIL or tensor - case insensitive")
    parser.add_argument("--use-v2", action="store_true", help="Use V2 transforms")
    parser.add_argument("--model-path", default=None, help="Path to model checkpoint.")
    return parser


if __name__ == "__main__":
    args = get_args_parser().parse_args()
    main(args)