lama

1
#!/usr/bin/env python3
2

3
import os
4

5
import pandas as pd
6

7
from saicinpainting.evaluation.data import PrecomputedInpaintingResultsDataset
8
from saicinpainting.evaluation.evaluator import InpaintingEvaluator, lpips_fid100_f1
9
from saicinpainting.evaluation.losses.base_loss import SegmentationAwareSSIM, \
10
    SegmentationClassStats, SSIMScore, LPIPSScore, FIDScore, SegmentationAwareLPIPS, SegmentationAwareFID
11
from saicinpainting.evaluation.utils import load_yaml
12

13

14
def main(args):
15
    config = load_yaml(args.config)
16

17
    dataset = PrecomputedInpaintingResultsDataset(args.datadir, args.predictdir, **config.dataset_kwargs)
18

19
    metrics = {
20
        'ssim': SSIMScore(),
21
        'lpips': LPIPSScore(),
22
        'fid': FIDScore()
23
    }
24
    enable_segm = config.get('segmentation', dict(enable=False)).get('enable', False)
25
    if enable_segm:
26
        weights_path = os.path.expandvars(config.segmentation.weights_path)
27
        metrics.update(dict(
28
            segm_stats=SegmentationClassStats(weights_path=weights_path),
29
            segm_ssim=SegmentationAwareSSIM(weights_path=weights_path),
30
            segm_lpips=SegmentationAwareLPIPS(weights_path=weights_path),
31
            segm_fid=SegmentationAwareFID(weights_path=weights_path)
32
        ))
33
    evaluator = InpaintingEvaluator(dataset, scores=metrics,
34
                                    integral_title='lpips_fid100_f1', integral_func=lpips_fid100_f1,
35
                                    **config.evaluator_kwargs)
36

37
    os.makedirs(os.path.dirname(args.outpath), exist_ok=True)
38

39
    results = evaluator.evaluate()
40

41
    results = pd.DataFrame(results).stack(1).unstack(0)
42
    results.dropna(axis=1, how='all', inplace=True)
43
    results.to_csv(args.outpath, sep='\t', float_format='%.4f')
44

45
    if enable_segm:
46
        only_short_results = results[[c for c in results.columns if not c[0].startswith('segm_')]].dropna(axis=1, how='all')
47
        only_short_results.to_csv(args.outpath + '_short', sep='\t', float_format='%.4f')
48

49
        print(only_short_results)
50

51
        segm_metrics_results = results[['segm_ssim', 'segm_lpips', 'segm_fid']].dropna(axis=1, how='all').transpose().unstack(0).reorder_levels([1, 0], axis=1)
52
        segm_metrics_results.drop(['mean', 'std'], axis=0, inplace=True)
53

54
        segm_stats_results = results['segm_stats'].dropna(axis=1, how='all').transpose()
55
        segm_stats_results.index = pd.MultiIndex.from_tuples(n.split('/') for n in segm_stats_results.index)
56
        segm_stats_results = segm_stats_results.unstack(0).reorder_levels([1, 0], axis=1)
57
        segm_stats_results.sort_index(axis=1, inplace=True)
58
        segm_stats_results.dropna(axis=0, how='all', inplace=True)
59

60
        segm_results = pd.concat([segm_metrics_results, segm_stats_results], axis=1, sort=True)
61
        segm_results.sort_values(('mask_freq', 'total'), ascending=False, inplace=True)
62

63
        segm_results.to_csv(args.outpath + '_segm', sep='\t', float_format='%.4f')
64
    else:
65
        print(results)
66

67

68
if __name__ == '__main__':
69
    import argparse
70

71
    aparser = argparse.ArgumentParser()
72
    aparser.add_argument('config', type=str, help='Path to evaluation config')
73
    aparser.add_argument('datadir', type=str,
74
                         help='Path to folder with images and masks (output of gen_mask_dataset.py)')
75
    aparser.add_argument('predictdir', type=str,
76
                         help='Path to folder with predicts (e.g. predict_hifill_baseline.py)')
77
    aparser.add_argument('outpath', type=str, help='Where to put results')
78

79
    main(aparser.parse_args())
80