forked from rotem-shalev/Ham2Pose
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtest.py
194 lines (159 loc) · 8.51 KB
/
test.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
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
import os
import json
import yaml
import numpy as np
import matplotlib.pyplot as plt
import torch
import sys
rootdir = os.path.dirname(os.path.dirname(os.path.realpath(__file__)))
sys.path.insert(0, rootdir)
from args import args
from data.data import get_dataset
from constants import DATASET_SIZE, num_steps_to_batch_size
from model import IterativeTextGuidedPoseGenerationModel
from train import get_model_args
from predict import pred, predict_pose
from metrics import get_poses_ranks
def combine_results(experiment_name, results_path):
results = dict()
for file in os.listdir(results_path):
if experiment_name in file:
with open(os.path.join(results_path, file)) as f:
results.update(json.load(f))
return np.mean(list(results.values())), np.median(list(results.values()))
def get_lang(sign_id):
if "pjm" in sign_id:
return "pjm"
elif "gsl" in sign_id:
return "gsl"
elif sign_id.isnumeric():
return "dgs"
else:
return "lsf"
def get_results_by_language(filename, num_files=5):
paths = [os.path.join(filename+f"_{i}.txt") for i in range(num_files)]
languages = {"pjm", "dgs", "gsl", "lsf"}
all_results = {lang: {"pred_rank1": 0, "pred_rank5": 0, "pred_rank10": 0, "gt_rank1": 0, "gt_rank5": 0,
"gt_rank10": 0} for lang in languages}
lang2count = {lang: 0 for lang in languages}
for path in paths:
with open(path, 'r') as f:
lines = f.readlines()
for i in range(1, len(lines), 2):
if lines[i].startswith("rank"):
break
lang = get_lang(lines[i].split(" ")[0])
lang2count[lang] += 1
dist, pred_rank1, pred_rank5, pred_rank10, gt_rank1, gt_rank5, gt_rank10 = lines[i+1].strip().split(", ")
all_results[lang]["pred_rank1"] += int(pred_rank1 == "True")
all_results[lang]["pred_rank5"] += int(pred_rank5 == "True")
all_results[lang]["pred_rank10"] += int(pred_rank10 == "True")
all_results[lang]["gt_rank1"] += int(gt_rank1 == "True")
all_results[lang]["gt_rank5"] += int(gt_rank5 == "True")
all_results[lang]["gt_rank10"] += int(gt_rank10 == "True")
for lang, ranks in all_results.items():
for rank, rank_count in ranks.items():
rank_mean = rank_count / lang2count[lang]
print(f"{rank} of {lang} is: {rank_count}/{lang2count[lang]}= {rank_mean}")
def test_seq_len(model, dataset, model_name):
abs_diffs = dict()
diffs = dict()
for d in dataset:
_, seq_len = model.encode_text([d["text"]])
real_seq_len = d["pose"]["length"]
diff = seq_len.item() - real_seq_len.item()
abs_diffs[d["id"]] = np.abs(diff)
diffs[d["id"]] = diff / real_seq_len.item()
print(f"mean diff: {np.mean(list(diffs.values()))}, median: {np.median(list(diffs.values()))}")
print(f"mean absolute diff: {np.mean(list(abs_diffs.values()))}, median:"
f" {np.median(list(abs_diffs.values()))}")
plt.hist([v * 100 for v in diffs.values()], bins=80)
plt.xticks(ticks=[-50, 0, 50, 100, 150], labels=["-50%", "0%", "50%", "100%", "150%"])
# plt.title("Predicted vs. real sequence length difference")
plt.xlabel('sequence length error percentage')
plt.ylabel('Count')
plt.savefig(f"models/{model_name}/results/seq_len_diff_hist.png")
plt.clf()
plt.hist(list(abs_diffs.values()), bins=10)
# plt.title("Predicted vs. real sequence length absolute difference")
plt.xlabel('frame number difference (FPS=25)')
plt.ylabel('Count')
plt.savefig(f"models/{model_name}/results/seq_len_abs_diff_hist.png")
plt.clf()
with open(f"models/{model_name}/results/seq_len_diffs.json", 'w') as f:
json.dump(diffs, f)
with open(f"models/{model_name}/results/seq_len_abs_diffs.json", 'w') as f:
json.dump(abs_diffs, f)
def test_distance_ranks(model, model_name, dataset, keypoints_path, num_samples=20):
keypoints_dirs = os.listdir(keypoints_path)
with open("data/hamnosys/data.json", 'r') as f:
data = json.load(f)
data_ids = list(filter(lambda x: x in keypoints_dirs, data.keys()))
model = model.cuda()
with torch.no_grad():
rank_1_pred_sum = rank_5_pred_sum = rank_10_pred_sum = rank_1_label_sum = rank_5_label_sum = \
rank_10_label_sum = 0
pred2label_distances = dict()
for datum in dataset:
if len(datum["pose"]["data"]) == 0:
continue
predicted_pose = predict_pose(model, datum, pose_header)
pred2label_distance, rank_1_pred, rank_5_pred, rank_10_pred, rank_1_label, rank_5_label, \
rank_10_label = get_poses_ranks(predicted_pose, datum["id"], keypoints_path, data_ids,
model=model, pose_header=pose_header, ds=dataset, num_samples=num_samples)
pred2label_distances[datum["id"]] = pred2label_distance
print(f"{datum['id']} ranks:\n "
f"{pred2label_distance}, {rank_1_pred}, {rank_5_pred}, {rank_10_pred}, {rank_1_label},"
f" {rank_5_label}, {rank_10_label}")
rank_1_pred_sum += int(rank_1_pred)
rank_5_pred_sum += int(rank_5_pred)
rank_10_pred_sum += int(rank_10_pred)
rank_1_label_sum += int(rank_1_label)
rank_5_label_sum += int(rank_5_label)
rank_10_label_sum += int(rank_10_label)
num_samples = len(dataset)
print(f"rank 1 pred sum: {rank_1_pred_sum} / {num_samples}: {rank_1_pred_sum / num_samples}")
print(f"rank 5 pred sum: {rank_5_pred_sum} / {num_samples}: {rank_5_pred_sum / num_samples}")
print(f"rank 10 pred sum: {rank_10_pred_sum} / {num_samples}: {rank_10_pred_sum / num_samples}")
print(f"rank 1 label sum: {rank_1_label_sum} / {num_samples}: {rank_1_label_sum / num_samples}")
print(f"rank 5 label sum: {rank_5_label_sum} / {num_samples}: {rank_5_label_sum / num_samples}")
print(f"rank 10 label sum: {rank_10_label_sum} / {num_samples}: {rank_10_label_sum / num_samples}")
with open(f"models/{model_name}/results/pred2label_distances_NDTW_pred_label_gallery.json", 'w') as f:
json.dump(pred2label_distances, f)
print(f"mean distance between pred and label: {np.mean(list(pred2label_distances.values()))}")
print(f"median distance between pred and label: {np.median(list(pred2label_distances.values()))}")
plt.hist(list(pred2label_distances.values()))
plt.title("DTW distance between ground truth and predicted pose")
plt.savefig(f"models/{model_name}/results/pred2label_distances_hist.png")
def test(model, model_name, dataset, test_seq_len_predictor=True, test_ranks=True, output_dir="",
keypoints_path=""):
os.makedirs(f"models/{model_name}/results", exist_ok=True)
if output_dir != "":
pred(model, dataset, f"models/{model_name}/{output_dir}")
if test_seq_len_predictor:
test_seq_len(model, dataset, model_name)
if test_ranks:
test_distance_ranks(model, model_name, dataset, keypoints_path)
if __name__ == "__main__":
args = vars(args)
if args["config_file"]: # override args with yaml config file
with open(args["config_file"], 'r') as f:
args = yaml.safe_load(f)
args["batch_size"] = num_steps_to_batch_size[args["num_steps"]]
test_size = int(0.1 * DATASET_SIZE)
if args["leave_out"] != "":
_, dataset = get_dataset(name=args["dataset"], poses=args["pose"], fps=args["fps"],
components=args["pose_components"], leave_out=args["leave_out"],
max_seq_size=args["max_seq_size"], split='test')
else:
dataset = get_dataset(name=args["dataset"], poses=args["pose"], fps=args["fps"],
components=args["pose_components"], max_seq_size=args["max_seq_size"],
split=f'test[:{test_size}]')
_, num_pose_joints, num_pose_dims = dataset[0]["pose"]["data"].shape
pose_header = dataset.data[0]["pose"].header
model_args = get_model_args(args, num_pose_joints, num_pose_dims)
ckpt = f"./models/{args['model_name']}/{args['ckpt']}/model.ckpt"
model = IterativeTextGuidedPoseGenerationModel.load_from_checkpoint(ckpt, **model_args)
model.eval()
test(model, args["model_name"], dataset, test_seq_len_predictor=True, test_ranks=False,
output_dir=args["output_dir"], keypoints_path="data/hamnosys/keypoints")