| |
| |
| |
| |
| |
| import warnings |
|
|
| warnings.filterwarnings("ignore", category=RuntimeWarning) |
| import os.path as osp |
| import time |
| import argparse |
| import json |
| import numpy as np |
| import multiprocessing as mp |
| import pandas as pd |
| from pycocotools import mask as cocomask |
|
|
| from third_parts.revos.utils.metircs import db_eval_iou, db_eval_boundary, get_r2vos_accuracy, get_r2vos_robustness |
|
|
| NUM_WOEKERS = 128 |
|
|
| def eval_queue(q, rank, out_dict): |
| while not q.empty(): |
| vid_name, exp = q.get() |
| vid = exp_dict[vid_name] |
| exp_name = f'{vid_name}_{exp}' |
|
|
| pred = mask_pred_dict[vid_name][exp] |
|
|
| vid_len, h, w = len(vid['frames']), vid['height'], vid['width'] |
| gt_masks = np.zeros((vid_len, h, w), dtype=np.uint8) |
| pred_masks = np.zeros((vid_len, h, w), dtype=np.uint8) |
|
|
| anno_ids = vid['expressions'][exp]['anno_id'] |
|
|
| for frame_idx, frame_name in enumerate(vid['frames']): |
| |
| for anno_id in anno_ids: |
| mask_rle = mask_dict[str(anno_id)][frame_idx] |
| if mask_rle: |
| gt_masks[frame_idx] += cocomask.decode(mask_rle) |
|
|
|
|
| pred_mask = cocomask.decode(pred['prediction_masks'][frame_idx]) |
| pred_masks[frame_idx] += pred_mask |
| j = db_eval_iou(gt_masks, pred_masks).mean() |
| f = db_eval_boundary(gt_masks, pred_masks).mean() |
| a = get_r2vos_accuracy(gt_masks, pred_masks).mean() |
|
|
| out_dict[exp_name] = [j, f, a] |
|
|
|
|
| if __name__ == '__main__': |
| parser = argparse.ArgumentParser() |
| parser.add_argument("pred_path", type=str, ) |
| parser.add_argument("--exp_path", type=str, default="data/video_datas/revos/meta_expressions_valid_.json") |
| parser.add_argument("--mask_path", type=str, default="data/video_datas/revos/mask_dict.json") |
| parser.add_argument("--save_json_name", type=str, default="revos_valid.json") |
| parser.add_argument("--save_csv_name", type=str, default="revos_valid.csv") |
| args = parser.parse_args() |
| queue = mp.Queue() |
| exp_dict = json.load(open(args.exp_path))['videos'] |
| mask_dict = json.load(open(args.mask_path)) |
|
|
| mask_pred = json.load(open(args.pred_path)) |
|
|
| shared_exp_dict = mp.Manager().dict(exp_dict) |
| shared_mask_dict = mp.Manager().dict(mask_dict) |
| output_dict = mp.Manager().dict() |
|
|
| mask_pred_dict = mp.Manager().dict(mask_pred) |
|
|
|
|
| for idx, vid_name in enumerate(exp_dict): |
| vid = exp_dict[vid_name] |
| for exp in vid['expressions']: |
| queue.put([vid_name, exp]) |
|
|
| start_time = time.time() |
| if NUM_WOEKERS > 1: |
| processes = [] |
| for rank in range(NUM_WOEKERS): |
| p = mp.Process(target=eval_queue, args=(queue, rank, output_dict)) |
| p.start() |
| processes.append(p) |
|
|
| for p in processes: |
| p.join() |
| else: |
| eval_queue(queue, 0, output_dict) |
|
|
| |
| output_json_path = osp.join(osp.dirname(args.pred_path), args.save_json_name) |
| output_csv_path = osp.join(osp.dirname(args.pred_path), args.save_csv_name) |
|
|
| data_list = [] |
| for videxp, (j, f, a) in output_dict.items(): |
| vid_name, exp = videxp.rsplit('_', 1) |
| data = {} |
|
|
| data['video_name'] = vid_name |
| data['exp_id'] = exp |
| data['exp'] = exp_dict[vid_name]['expressions'][exp]['exp'] |
| data['videxp'] = videxp |
| data['J'] = round(100 * j, 2) |
| data['F'] = round(100 * f, 2) |
| data['JF'] = round(100 * (j + f) / 2, 2) |
| data['A'] = round(100 * a, 2) |
| data['type_id'] = exp_dict[vid_name]['expressions'][exp]['type_id'] |
|
|
| data_list.append(data) |
|
|
| is_long = lambda x: x['type_id'] == 0 |
| is_short = lambda x: x['type_id'] == 1 |
|
|
| j_referring = np.array([d['J'] for d in data_list if is_long(d)]).mean() |
| f_referring = np.array([d['F'] for d in data_list if is_long(d)]).mean() |
| a_referring = np.array([d['A'] for d in data_list if is_long(d)]).mean() |
| jf_referring = (j_referring + f_referring) / 2 |
|
|
| j_reason = np.array([d['J'] for d in data_list if is_short(d)]).mean() |
| f_reason = np.array([d['F'] for d in data_list if is_short(d)]).mean() |
| a_reason = np.array([d['A'] for d in data_list if is_short(d)]).mean() |
| jf_reason = (j_reason + f_reason) / 2 |
|
|
| j_referring_reason = (j_referring + j_reason) / 2 |
| f_referring_reason = (f_referring + f_reason) / 2 |
| a_referring_reason = (a_referring + a_reason) / 2 |
| jf_referring_reason = (jf_referring + jf_reason) / 2 |
|
|
| results = { |
| "long": { |
| "J" : j_referring, |
| "F" : f_referring, |
| "A" : a_referring, |
| "JF": jf_referring |
| }, |
| "short": { |
| "J" : j_reason, |
| "F" : f_reason, |
| "A" : a_reason, |
| "JF": jf_reason |
| }, |
| "overall": { |
| "J" : j_referring_reason, |
| "F" : f_referring_reason, |
| "A" : a_referring_reason, |
| "JF": jf_referring_reason |
| } |
| } |
|
|
| print(results) |
| with open(output_json_path, 'w') as f: |
| json.dump(results, f, indent=4) |
| print(f"Results saved to {output_json_path}") |
|
|
| data4csv = {} |
| for data in data_list: |
| for k, v in data.items(): |
| data4csv[k] = data4csv.get(k, []) + [v] |
|
|
| df = pd.DataFrame(data4csv) |
| df.to_csv(output_csv_path, index=False) |
| print(f"Results saved to {output_csv_path}") |
|
|
| end_time = time.time() |
| total_time = end_time - start_time |
| print("time: %.4f s" %(total_time)) |
|
|
