mmaction/evaluation/functional/ava_utils.py

# Copyright (c) OpenMMLab. All rights reserved.
# This piece of code is directly adapted from ActivityNet official repo
# https://github.com/activitynet/ActivityNet/blob/master/
# Evaluation/get_ava_performance.py. Some unused codes are removed.
import csv
import multiprocessing
import time
from collections import defaultdict

import numpy as np

from .ava_evaluation import metrics, np_box_list, np_box_ops


def det2csv(results, custom_classes):
    """Convert detection results to csv file."""
    csv_results = []
    for idx in range(len(results)):
        video_id = results[idx]['video_id']
        timestamp = results[idx]['timestamp']
        result = results[idx]['outputs']
        for label, _ in enumerate(result):
            for bbox in result[label]:
                bbox_ = tuple(bbox.tolist())
                if custom_classes is not None:
                    actual_label = custom_classes[label + 1]
                else:
                    actual_label = label + 1
                csv_results.append((
                    video_id,
                    timestamp,
                ) + bbox_[:4] + (actual_label, ) + bbox_[4:])
    return csv_results


# results is organized by class
def results2csv(results, out_file, custom_classes=None):
    """Convert detection results to csv file."""
    csv_results = det2csv(results, custom_classes)

    # save space for float
    def to_str(item):
        if isinstance(item, float):
            return f'{item:.4f}'
        return str(item)

    with open(out_file, 'w') as f:
        for csv_result in csv_results:
            f.write(','.join(map(to_str, csv_result)))
            f.write('\n')


def print_time(message, start):
    """Print processing time."""
    print('==> %g seconds to %s' % (time.time() - start, message), flush=True)


def make_image_key(video_id, timestamp):
    """Returns a unique identifier for a video id & timestamp."""
    return f'{video_id},{int(timestamp):04d}'


def read_csv(csv_file, class_whitelist=None):
    """Loads boxes and class labels from a CSV file in the AVA format.

    CSV file format described at https://research.google.com/ava/download.html.

    Args:
        csv_file: A file object.
        class_whitelist: If provided, boxes corresponding to (integer) class
        labels not in this set are skipped.

    Returns:
        boxes: A dictionary mapping each unique image key (string) to a list of
        boxes, given as coordinates [y1, x1, y2, x2].
        labels: A dictionary mapping each unique image key (string) to a list
        of integer class labels, matching the corresponding box in `boxes`.
        scores: A dictionary mapping each unique image key (string) to a list
        of score values labels, matching the corresponding label in `labels`.
        If scores are not provided in the csv, then they will default to 1.0.
    """
    entries = defaultdict(list)
    boxes = defaultdict(list)
    labels = defaultdict(list)
    scores = defaultdict(list)
    reader = csv.reader(csv_file)
    for row in reader:
        assert len(row) in [7, 8], 'Wrong number of columns: ' + row
        image_key = make_image_key(row[0], row[1])
        x1, y1, x2, y2 = [float(n) for n in row[2:6]]
        action_id = int(row[6])
        if class_whitelist and action_id not in class_whitelist:
            continue

        score = 1.0
        if len(row) == 8:
            score = float(row[7])

        entries[image_key].append((score, action_id, y1, x1, y2, x2))

    for image_key in entries:
        # Evaluation API assumes boxes with descending scores
        entry = sorted(entries[image_key], key=lambda tup: -tup[0])
        boxes[image_key] = [x[2:] for x in entry]
        labels[image_key] = [x[1] for x in entry]
        scores[image_key] = [x[0] for x in entry]

    return boxes, labels, scores


def read_exclusions(exclusions_file):
    """Reads a CSV file of excluded timestamps.

    Args:
        exclusions_file: A file object containing a csv of video-id,timestamp.

    Returns:
        A set of strings containing excluded image keys, e.g.
        "aaaaaaaaaaa,0904",
        or an empty set if exclusions file is None.
    """
    excluded = set()
    if exclusions_file:
        reader = csv.reader(exclusions_file)
    for row in reader:
        assert len(row) == 2, f'Expected only 2 columns, got: {row}'
        excluded.add(make_image_key(row[0], row[1]))
    return excluded


def read_labelmap(labelmap_file):
    """Reads a labelmap without the dependency on protocol buffers.

    Args:
        labelmap_file: A file object containing a label map protocol buffer.

    Returns:
        labelmap: The label map in the form used by the
        object_detection_evaluation
        module - a list of {"id": integer, "name": classname } dicts.
        class_ids: A set containing all of the valid class id integers.
    """
    labelmap = []
    class_ids = set()
    name = ''
    class_id = ''
    for line in labelmap_file:
        if line.startswith('  name:'):
            name = line.split('"')[1]
        elif line.startswith('  id:') or line.startswith('  label_id:'):
            class_id = int(line.strip().split(' ')[-1])
            labelmap.append({'id': class_id, 'name': name})
            class_ids.add(class_id)
    return labelmap, class_ids


def get_overlaps_and_scores_box_mode(detected_boxes, detected_scores,
                                     groundtruth_boxes):

    detected_boxlist = np_box_list.BoxList(detected_boxes)
    detected_boxlist.add_field('scores', detected_scores)
    gt_non_group_of_boxlist = np_box_list.BoxList(groundtruth_boxes)

    iou = np_box_ops.iou(detected_boxlist.get(), gt_non_group_of_boxlist.get())
    scores = detected_boxlist.get_field('scores')
    num_boxes = detected_boxlist.num_boxes()
    return iou, scores, num_boxes


def tpfp_single(tup, threshold=0.5):
    gt_bboxes, gt_labels, bboxes, labels, scores = tup
    ret_scores, ret_tp_fp_labels = dict(), dict()
    all_labels = list(set(labels))
    for label in all_labels:
        gt_bbox = np.array(
            [x for x, y in zip(gt_bboxes, gt_labels) if y == label],
            dtype=np.float32).reshape(-1, 4)
        bbox = np.array([x for x, y in zip(bboxes, labels) if y == label],
                        dtype=np.float32).reshape(-1, 4)
        score = np.array([x for x, y in zip(scores, labels) if y == label],
                         dtype=np.float32).reshape(-1)
        iou, score, num_boxes = get_overlaps_and_scores_box_mode(
            bbox, score, gt_bbox)
        if gt_bbox.size == 0:
            ret_scores[label] = score
            ret_tp_fp_labels[label] = np.zeros(num_boxes, dtype=bool)
            continue
        tp_fp_labels = np.zeros(num_boxes, dtype=bool)
        if iou.shape[1] > 0:
            max_overlap_gt_ids = np.argmax(iou, axis=1)
            is_gt_box_detected = np.zeros(iou.shape[1], dtype=bool)
            for i in range(num_boxes):
                gt_id = max_overlap_gt_ids[i]
                if iou[i, gt_id] >= threshold:
                    if not is_gt_box_detected[gt_id]:
                        tp_fp_labels[i] = True
                        is_gt_box_detected[gt_id] = True
        ret_scores[label], ret_tp_fp_labels[label] = score, tp_fp_labels
    return ret_scores, ret_tp_fp_labels


# Seems there is at most 100 detections for each image
def ava_eval(result_file,
             result_type,
             label_file,
             ann_file,
             exclude_file,
             verbose=True,
             ignore_empty_frames=True,
             custom_classes=None):
    """Perform ava evaluation."""

    assert result_type in ['mAP']
    start = time.time()
    categories, class_whitelist = read_labelmap(open(label_file))
    if custom_classes is not None:
        custom_classes = custom_classes[1:]
        assert set(custom_classes).issubset(set(class_whitelist))
        class_whitelist = custom_classes
        categories = [cat for cat in categories if cat['id'] in custom_classes]

    # loading gt, do not need gt score
    gt_bboxes, gt_labels, _ = read_csv(open(ann_file), class_whitelist)
    if verbose:
        print_time('Reading GT results', start)

    if exclude_file is not None:
        excluded_keys = read_exclusions(open(exclude_file))
    else:
        excluded_keys = list()

    start = time.time()
    boxes, labels, scores = read_csv(open(result_file), class_whitelist)
    if verbose:
        print_time('Reading Detection results', start)

    start = time.time()
    all_gt_labels = np.concatenate(list(gt_labels.values()))
    gt_count = {k: np.sum(all_gt_labels == k) for k in class_whitelist}

    pool = multiprocessing.Pool(32)
    if ignore_empty_frames:
        tups = [(gt_bboxes[k], gt_labels[k], boxes[k], labels[k], scores[k])
                for k in gt_bboxes if k not in excluded_keys]
    else:
        tups = [(gt_bboxes.get(k, np.zeros((0, 4), dtype=np.float32)),
                 gt_labels.get(k, []), boxes[k], labels[k], scores[k])
                for k in boxes if k not in excluded_keys]
    rets = pool.map(tpfp_single, tups)

    if verbose:
        print_time('Calculating TP/FP', start)

    start = time.time()
    scores, tpfps = defaultdict(list), defaultdict(list)
    for score, tpfp in rets:
        for k in score:
            scores[k].append(score[k])
            tpfps[k].append(tpfp[k])

    cls_AP = []
    for k in scores:
        scores[k] = np.concatenate(scores[k])
        tpfps[k] = np.concatenate(tpfps[k])
        precision, recall = metrics.compute_precision_recall(
            scores[k], tpfps[k], gt_count[k])
        ap = metrics.compute_average_precision(precision, recall)
        class_name = [x['name'] for x in categories if x['id'] == k]
        assert len(class_name) == 1
        class_name = class_name[0]
        cls_AP.append((k, class_name, ap))
    if verbose:
        print_time('Run Evaluator', start)

    print('Per-class results: ', flush=True)
    for k, class_name, ap in cls_AP:
        print(f'Index: {k}, Action: {class_name}: AP: {ap:.4f};', flush=True)

    overall = np.nanmean([x[2] for x in cls_AP])
    person_movement = np.nanmean([x[2] for x in cls_AP if x[0] <= 14])
    object_manipulation = np.nanmean([x[2] for x in cls_AP if 14 < x[0] < 64])
    person_interaction = np.nanmean([x[2] for x in cls_AP if 64 <= x[0]])

    print('Overall Results: ', flush=True)
    print(f'Overall mAP: {overall:.4f}', flush=True)
    print(f'Person Movement mAP: {person_movement:.4f}', flush=True)
    print(f'Object Manipulation mAP: {object_manipulation:.4f}', flush=True)
    print(f'Person Interaction mAP: {person_interaction:.4f}', flush=True)

    results = {}
    results['overall'] = overall
    results['person_movement'] = person_movement
    results['object_manipulation'] = object_manipulation
    results['person_interaction'] = person_interaction

    if verbose:
        for k, class_name, ap in cls_AP:
            print(f'Class {class_name} AP: {ap:.4f}', flush=True)

    return results