coco_utils.py

# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved
"""
COCO evaluator that works in distributed mode.
Mostly copy-paste from https://github.com/pytorch/vision/blob/edfd5a7/references/detection/coco_eval.py
The difference is that there is less copy-pasting from pycocotools
in the end of the file, as python3 can suppress prints with contextlib
"""
import os
import contextlib
import copy
import numpy as np
import torch
import torchvision
import torch.distributed as dist

from pycocotools.cocoeval import COCOeval
from pycocotools.coco import COCO
import pycocotools.mask as mask_util

import pickle

def is_dist_avail_and_initialized():
    if not dist.is_available():
        return False
    if not dist.is_initialized():
        return False
    return True


def get_world_size():
    if not is_dist_avail_and_initialized():
        return 1
    return dist.get_world_size()

def all_gather(data):
    """
    Run all_gather on arbitrary picklable data (not necessarily tensors)
    Args:
        data: any picklable object
    Returns:
        list[data]: list of data gathered from each rank
    """
    world_size = get_world_size()
    if world_size == 1:
        return [data]

    # serialized to a Tensor
    buffer = pickle.dumps(data)
    storage = torch.ByteStorage.from_buffer(buffer)
    tensor = torch.ByteTensor(storage).to("cuda")

    # obtain Tensor size of each rank
    local_size = torch.tensor([tensor.numel()], device="cuda")
    size_list = [torch.tensor([0], device="cuda") for _ in range(world_size)]
    dist.all_gather(size_list, local_size)
    size_list = [int(size.item()) for size in size_list]
    max_size = max(size_list)

    # receiving Tensor from all ranks
    # we pad the tensor because torch all_gather does not support
    # gathering tensors of different shapes
    tensor_list = []
    for _ in size_list:
        tensor_list.append(torch.empty((max_size,), dtype=torch.uint8, device="cuda"))
    if local_size != max_size:
        padding = torch.empty(size=(max_size - local_size,), dtype=torch.uint8, device="cuda")
        tensor = torch.cat((tensor, padding), dim=0)
    dist.all_gather(tensor_list, tensor)

    data_list = []
    for size, tensor in zip(size_list, tensor_list):
        buffer = tensor.cpu().numpy().tobytes()[:size]
        data_list.append(pickle.loads(buffer))

    return data_list


def get_coco_api_from_dataset(dataset):
    for _ in range(10):
        # if isinstance(dataset, torchvision.datasets.CocoDetection):
        #     break
        if isinstance(dataset, torch.utils.data.Subset):
            dataset = dataset.dataset
    if isinstance(dataset, torchvision.datasets.CocoDetection):
        return dataset.coco


class CocoEvaluator(object):
    def __init__(self, coco_gt, iou_types):
        assert isinstance(iou_types, (list, tuple))
        self.coco_gt = copy.deepcopy(coco_gt)

        self.iou_types = iou_types
        self.coco_eval = {}
        for iou_type in iou_types:
            self.coco_eval[iou_type] = COCOeval(self.coco_gt, iouType=iou_type)

        self.img_ids = []
        self.eval_imgs = {k: [] for k in iou_types}

    def update(self, predictions):
        img_ids = list(np.unique(list(predictions.keys())))
        self.img_ids.extend(img_ids)

        for iou_type in self.iou_types:
            results = self.prepare(predictions, iou_type)

            # suppress pycocotools prints
            with open(os.devnull, 'w') as devnull:
                with contextlib.redirect_stdout(devnull):
                    coco_dt = COCO.loadRes(self.coco_gt, results) if results else COCO()
            coco_eval = self.coco_eval[iou_type]

            coco_eval.cocoDt = coco_dt
            coco_eval.params.imgIds = list(img_ids)
            img_ids, eval_imgs = evaluate(coco_eval)

            self.eval_imgs[iou_type].append(eval_imgs)

    def synchronize_between_processes(self):
        for iou_type in self.iou_types:
            self.eval_imgs[iou_type] = np.concatenate(self.eval_imgs[iou_type], 2)
            create_common_coco_eval(self.coco_eval[iou_type], self.img_ids, self.eval_imgs[iou_type])

    def accumulate(self):
        for coco_eval in self.coco_eval.values():
            coco_eval.accumulate()

    def summarize(self):
        for iou_type, coco_eval in self.coco_eval.items():
            print("IoU metric: {}".format(iou_type))
            coco_eval.summarize()
    
    def _post_process_stats(self, stats, coco_eval_object, iou_type='bbox'):
        # bbox & segm:
        # stats[0] = _summarize(1)
        # stats[1] = _summarize(1, iouThr=.5, maxDets=self.params.maxDets[2])
        # stats[2] = _summarize(1, iouThr=.75, maxDets=self.params.maxDets[2])
        # stats[3] = _summarize(1, areaRng='small', maxDets=self.params.maxDets[2])
        # stats[4] = _summarize(1, areaRng='medium', maxDets=self.params.maxDets[2])
        # stats[5] = _summarize(1, areaRng='large', maxDets=self.params.maxDets[2])
        # stats[6] = _summarize(0, maxDets=self.params.maxDets[0])
        # stats[7] = _summarize(0, maxDets=self.params.maxDets[1])
        # stats[8] = _summarize(0, maxDets=self.params.maxDets[2])
        # stats[9] = _summarize(0, areaRng='small', maxDets=self.params.maxDets[2])
        # stats[10] = _summarize(0, areaRng='medium', maxDets=self.params.maxDets[2])
        # stats[11] = _summarize(0, areaRng='large', maxDets=self.params.maxDets[2])

        # keypoints:
        # stats[0] = _summarize(1, maxDets=20)
        # stats[1] = _summarize(1, maxDets=20, iouThr=.5)
        # stats[2] = _summarize(1, maxDets=20, iouThr=.75)
        # stats[3] = _summarize(1, maxDets=20, areaRng='medium')
        # stats[4] = _summarize(1, maxDets=20, areaRng='large')
        # stats[5] = _summarize(0, maxDets=20)
        # stats[6] = _summarize(0, maxDets=20, iouThr=.5)
        # stats[7] = _summarize(0, maxDets=20, iouThr=.75)
        # stats[8] = _summarize(0, maxDets=20, areaRng='medium')
        # stats[9] = _summarize(0, maxDets=20, areaRng='large')
        if iou_type not in ['bbox', 'segm', 'keypoints']:
            raise ValueError(f"iou_type '{iou_type}' not supported")

        current_max_dets = coco_eval_object.params.maxDets

        index_to_title = {
            "bbox": {
                0: f"AP-IoU=0.50:0.95-area=all-maxDets={current_max_dets[2]}",
                1: f"AP-IoU=0.50-area=all-maxDets={current_max_dets[2]}",
                2: f"AP-IoU=0.75-area=all-maxDets={current_max_dets[2]}",
                3: f"AP-IoU=0.50:0.95-area=small-maxDets={current_max_dets[2]}",
                4: f"AP-IoU=0.50:0.95-area=medium-maxDets={current_max_dets[2]}",
                5: f"AP-IoU=0.50:0.95-area=large-maxDets={current_max_dets[2]}",
                6: f"AR-IoU=0.50:0.95-area=all-maxDets={current_max_dets[0]}",
                7: f"AR-IoU=0.50:0.95-area=all-maxDets={current_max_dets[1]}",
                8: f"AR-IoU=0.50:0.95-area=all-maxDets={current_max_dets[2]}",
                9: f"AR-IoU=0.50:0.95-area=small-maxDets={current_max_dets[2]}",
                10: f"AR-IoU=0.50:0.95-area=medium-maxDets={current_max_dets[2]}",
                11: f"AR-IoU=0.50:0.95-area=large-maxDets={current_max_dets[2]}",
            },
            "keypoints":
                {
                    0: "AP-IoU=0.50:0.95-area=all-maxDets=20",
                    1: "AP-IoU=0.50-area=all-maxDets=20",
                    2: "AP-IoU=0.75-area=all-maxDets=20",
                    3: "AP-IoU=0.50:0.95-area=medium-maxDets=20",
                    4: "AP-IoU=0.50:0.95-area=large-maxDets=20",
                    5: "AR-IoU=0.50:0.95-area=all-maxDets=20",
                    6: "AR-IoU=0.50-area=all-maxDets=20",
                    7: "AR-IoU=0.75-area=all-maxDets=20",
                    8: "AR-IoU=0.50:0.95-area=medium-maxDets=20",
                    9: "AR-IoU=0.50:0.95-area=large-maxDets=20",
                },
        }

        output_dict = {}
        for index, stat in enumerate(stats):
            output_dict[index_to_title[iou_type][index]] = stat
        
        return output_dict


    def get_results(self):
        output_dict = {}

        for iou_type, coco_eval in self.coco_eval.items():
            if iou_type == 'segm':
                iou_type = 'bbox'
            output_dict[f"iou_{iou_type}"] = self._post_process_stats(coco_eval.stats, coco_eval, iou_type)
        return output_dict

    def prepare(self, predictions, iou_type):
        if iou_type == "bbox":
            return self.prepare_for_coco_detection(predictions)
        elif iou_type == "segm":
            return self.prepare_for_coco_segmentation(predictions)
        elif iou_type == "keypoints":
            return self.prepare_for_coco_keypoint(predictions)
        else:
            raise ValueError("Unknown iou type {}".format(iou_type))

    def prepare_for_coco_detection(self, predictions):
        coco_results = []
        for original_id, prediction in predictions.items():
            if len(prediction) == 0:
                continue

            boxes = prediction["boxes"]
            if not isinstance(boxes, torch.Tensor):
                boxes = torch.as_tensor(boxes)
            boxes = convert_to_xywh(boxes).tolist()

            scores = prediction["scores"]
            if not isinstance(scores, list):
                scores = scores.tolist()

            labels = prediction["labels"]
            if not isinstance(labels, list):
                labels = prediction["labels"].tolist()

            coco_results.extend(
                [
                    {
                        "image_id": original_id,
                        "category_id": labels[k],
                        "bbox": box,
                        "score": scores[k],
                    }
                    for k, box in enumerate(boxes)
                ]
            )
        return coco_results

    def prepare_for_coco_segmentation(self, predictions):
        coco_results = []
        for original_id, prediction in predictions.items():
            if len(prediction) == 0:
                continue

            scores = prediction["scores"]
            labels = prediction["labels"]
            masks = prediction["masks"]

            masks = masks > 0.5

            scores = prediction["scores"].tolist()
            labels = prediction["labels"].tolist()

            rles = [
                mask_util.encode(np.array(mask[0, :, :, np.newaxis], dtype=np.uint8, order="F"))[0]
                for mask in masks
            ]
            for rle in rles:
                rle["counts"] = rle["counts"].decode("utf-8")

            coco_results.extend(
                [
                    {
                        "image_id": original_id,
                        "category_id": labels[k],
                        "segmentation": rle,
                        "score": scores[k],
                    }
                    for k, rle in enumerate(rles)
                ]
            )
        return coco_results

    def prepare_for_coco_keypoint(self, predictions):
        coco_results = []
        for original_id, prediction in predictions.items():
            if len(prediction) == 0:
                continue

            boxes = prediction["boxes"]
            boxes = convert_to_xywh(boxes).tolist()
            scores = prediction["scores"].tolist()
            labels = prediction["labels"].tolist()
            keypoints = prediction["keypoints"]
            keypoints = keypoints.flatten(start_dim=1).tolist()

            coco_results.extend(
                [
                    {
                        "image_id": original_id,
                        "category_id": labels[k],
                        'keypoints': keypoint,
                        "score": scores[k],
                    }
                    for k, keypoint in enumerate(keypoints)
                ]
            )
        return coco_results


def convert_to_xywh(boxes):
    xmin, ymin, xmax, ymax = boxes.unbind(1)
    return torch.stack((xmin, ymin, xmax - xmin, ymax - ymin), dim=1)


def merge(img_ids, eval_imgs):
    all_img_ids = all_gather(img_ids)
    all_eval_imgs = all_gather(eval_imgs)

    merged_img_ids = []
    for p in all_img_ids:
        merged_img_ids.extend(p)

    merged_eval_imgs = []
    for p in all_eval_imgs:
        merged_eval_imgs.append(p)

    merged_img_ids = np.array(merged_img_ids)
    merged_eval_imgs = np.concatenate(merged_eval_imgs, 2)

    # keep only unique (and in sorted order) images
    merged_img_ids, idx = np.unique(merged_img_ids, return_index=True)
    merged_eval_imgs = merged_eval_imgs[..., idx]

    return merged_img_ids, merged_eval_imgs


def create_common_coco_eval(coco_eval, img_ids, eval_imgs):
    img_ids, eval_imgs = merge(img_ids, eval_imgs)
    img_ids = list(img_ids)
    eval_imgs = list(eval_imgs.flatten())

    coco_eval.evalImgs = eval_imgs
    coco_eval.params.imgIds = img_ids
    coco_eval._paramsEval = copy.deepcopy(coco_eval.params)


#################################################################
# From pycocotools, just removed the prints and fixed
# a Python3 bug about unicode not defined
#################################################################

def evaluate(self):
    '''
    Run per image evaluation on given images and store results (a list of dict) in self.evalImgs
    :return: None
    '''
    # tic = time.time()
    # print('Running per image evaluation...')
    p = self.params
    # add backward compatibility if useSegm is specified in params
    if p.useSegm is not None:
        p.iouType = 'segm' if p.useSegm == 1 else 'bbox'
        print('useSegm (deprecated) is not None. Running {} evaluation'.format(p.iouType))
    # print('Evaluate annotation type *{}*'.format(p.iouType))
    p.imgIds = list(np.unique(p.imgIds))
    if p.useCats:
        p.catIds = list(np.unique(p.catIds))
    p.maxDets = sorted(p.maxDets)
    self.params = p

    self._prepare()
    # loop through images, area range, max detection number
    catIds = p.catIds if p.useCats else [-1]

    if p.iouType == 'segm' or p.iouType == 'bbox':
        computeIoU = self.computeIoU
    elif p.iouType == 'keypoints':
        computeIoU = self.computeOks
    self.ious = {
        (imgId, catId): computeIoU(imgId, catId)
        for imgId in p.imgIds
        for catId in catIds}

    evaluateImg = self.evaluateImg
    maxDet = p.maxDets[-1]
    evalImgs = [
        evaluateImg(imgId, catId, areaRng, maxDet)
        for catId in catIds
        for areaRng in p.areaRng
        for imgId in p.imgIds
    ]
    # this is NOT in the pycocotools code, but could be done outside
    evalImgs = np.asarray(evalImgs).reshape(len(catIds), len(p.areaRng), len(p.imgIds))
    self._paramsEval = copy.deepcopy(self.params)
    # toc = time.time()
    # print('DONE (t={:0.2f}s).'.format(toc-tic))
    return p.imgIds, evalImgs