forked from naoto0804/pytorch-inpainting-with-partial-conv
-
Notifications
You must be signed in to change notification settings - Fork 0
/
scores.py
executable file
·161 lines (124 loc) · 4.88 KB
/
scores.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
#!/usr/bin/env python
# coding: utf-8
import os, sys, inspect
currentdir = os.path.dirname(os.path.abspath(inspect.getfile(inspect.currentframe())))
parentdir = os.path.dirname(currentdir)
sys.path.insert(0,parentdir)
import numpy as np
import pandas as pd
import pickle as pkl
import random
import torch
import torch.nn
import torch.nn.functional as F
from torch.utils.data import Dataset, DataLoader
from torchvision import transforms
import os
from PIL import Image
import util.opt as opt
from util.data import *
from network.net import PConvUNet
from util.image import unnormalize
from collections import OrderedDict
import tensorflow as tf
from tqdm import tqdm
import argparse
# arg parser
parser = argparse.ArgumentParser()
parser.add_argument('--model', required=True, help='name of the model to be evaluated')
args = parser.parse_args()
# ## Read masks
test_csv = pd.read_csv("./dataset/test.csv")
files = test_csv.to_numpy().reshape(-1,)
masks_paths = ["./dataset/masks/" + file + ".png" for file in files]
masks = [np.array(Image.open(mask_path)) for mask_path in masks_paths]
# compute fill ratio
hole2image = [np.sum(mask == 0) / np.prod(mask.shape) for mask in masks]
bins = [0.01, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6]
disc_hole2image = pd.cut(x=hole2image, bins=bins)
disc_hole2image = [str(x) for x in disc_hole2image]
dict_masks = {}
for i in range(len(masks)):
dh2i = disc_hole2image[i]
if dh2i == 'nan':
continue
dict_masks[dh2i] = dict_masks.get(dh2i, []) + [masks[i]]
ranges = ['(0.01, 0.1]', '(0.1, 0.2]', '(0.2, 0.3]', '(0.3, 0.4]', '(0.4, 0.5]', '(0.5, 0.6]']
values = [len(dict_masks[rng]) for rng in ranges]
# ## Compute scores
def score_func(gts: list, outputs: list, func_name:str):
scores = []
sess = tf.Session()
with sess.as_default():
for i in range(len(gts)):
gt, output = gts[i], outputs[i]
num_pixels = float(np.prod(gt.shape))
gt = tf.convert_to_tensor(gt)
output = tf.convert_to_tensor(output)
if func_name == 'PSNR':
score = tf.image.psnr(gt, output, max_val=1.0)
elif func_name == 'SSIM':
score = tf.image.ssim(gt, output, max_val=1.0)
else:
score = tf.divide(tf.reduce_sum(tf.abs(gt - output)), num_pixels / 255.)
score = score.eval()
if score == np.inf:
continue
scores.append(score)
return scores
# dictionary of metric functions
metric_funcs = [ "L1", "PSNR", "SSIM"]
# dictionary of metric scores
metric_scores = {
'(0.01, 0.1]': OrderedDict(),
'(0.1, 0.2]': OrderedDict(),
'(0.2, 0.3]': OrderedDict(),
'(0.3, 0.4]': OrderedDict(),
'(0.4, 0.5]': OrderedDict(),
'(0.5, 0.6]': OrderedDict()
}
# define device
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
# create dataset
dataset = NusceneDataset('./dataset', train=False)
dataloader = DataLoader(dataset, batch_size=32, num_workers=4, drop_last=False)
# load model
model = PConvUNet().to(device)
ckpt_dict = torch.load("./snapshots/ckpt/%s" % (args.model))
model.load_state_dict(ckpt_dict["model"], strict=False)
model.eval()
for rng in tqdm(dict_masks):
for i, sample in enumerate(dataloader):#len(dataset)):
gts = sample["gt"].float().to(device)
# sample a random maks from the current range
idx = np.random.choice(np.arange(len(dict_masks[rng])), size=gts.shape[0], replace=True)
masks = [dict_masks[rng][i] for i in idx]
masks = [np.expand_dims(mask / 255., 0).repeat(3, axis=0) for mask in masks]
masks = torch.tensor(np.stack(masks)).float().to(device)
# image masked
imgs = gts * masks
# get output
with torch.no_grad():
outputs, _ = model(imgs, masks)
outputs_comp = masks * imgs + (1 - masks) * outputs
# send back to cpu
gts = gts.cpu()
outputs_comp = outputs_comp.cpu()
# unnormalize
un_gts = unnormalize(gts)
un_outputs_comp = unnormalize(outputs_comp)
# get them to range 0, 1
un_gts = torch.clamp(un_gts, 0., 1.)
un_outputs_comp = torch.clamp(un_outputs_comp, 0., 1.)
# reshape and typing
un_gts = [x.transpose(1, 2, 0) for x in list(un_gts.numpy())]
un_outputs_comp = [x.transpose(1, 2, 0) for x in list(un_outputs_comp.numpy())]
for score_name in metric_funcs:
metric_scores[rng][score_name] = metric_scores[rng].get(score_name, []) + score_func(un_gts, un_outputs_comp, score_name)
statistics = {}
for rng in ranges:
statistics[rng] = []
for score_name in metric_funcs:
statistics[rng].append(np.mean(metric_scores[rng][score_name]))
results = pd.DataFrame(statistics, index=metric_funcs)
results.to_csv("../results/%s_results.csv" % (args.model))