mq_detect_performance.py

import os
from utils.eval_detection import ANETdetection

def run_evaluation(ground_truth_filename, prediction_filename,
         subset='test', tiou_thresholds=[0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7],
         verbose=True):

    anet_detection = ANETdetection(ground_truth_filename, prediction_filename,
                                   subset=subset, tiou_thresholds=tiou_thresholds,
                                   verbose=verbose, check_status=False)
    mAPs, average_mAP = anet_detection.evaluate()

    for (tiou, mAP) in zip(tiou_thresholds, mAPs):
        print("mAP at tIoU {} is {}".format(tiou, mAP))



def evaluation_detection(opt):

    # run_evaluation(ground_truth_filename = opt["clip_anno"],
    #                prediction_filename = os.path.join(opt["output_path"], opt["detect_result_file"]),
    #                subset=opt['infer_datasplit'], tiou_thresholds=opt['tIoU_thr'])
    run_evaluation(ground_truth_filename = opt["clip_anno"],
                prediction_filename = os.path.join(opt["output_path"], opt["detect_result_file"]),
                subset=opt['infer_datasplit'], tiou_thresholds=opt['tIoU_thr'])


if __name__ == "__main__":
    opt = {}
    opt["clip_anno"] = r'/work/sreeragiyer_umass_edu/ego4d_data/pkl_obj/sample_mq_val.pkl'
    opt["output_path"] = r'/work/sreeragiyer_umass_edu/ego4d/output/records/meme_mq_corr_waudio2004'
    opt["detect_result_file"] = "records_11.json"
    opt["infer_datasplit"] = "test"
    opt['tIoU_thr'] = [0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7]
    evaluation_detection(opt)