-
Notifications
You must be signed in to change notification settings - Fork 0
/
test.py
190 lines (170 loc) · 8.36 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
import torch
import torch.utils.data as data
from torch.utils.data import DataLoader
import numpy as np
import os
import json
import sys
from pytorch3d.io import load_obj
from pytorch3d.structures import Meshes
from pytorch3d.loss import chamfer_distance
from pytorch3d.ops import sample_points_from_meshes
from utils import AverageValueMeter, get_f_score
from external.PyTorchEMD.emd import earth_mover_distance
from chamfer_distance import ChamferDistance as chamfer_dist
distChamfer = chamfer_dist()
CD_NUM = 100000
EMD_NUM = 2048
REMOVE_OUTLIERS = True
REMOVE_OUTLIERS_THRESHOLD = 0.01
WORK_DIR = "./experiments/shapenet_cube_offset_generalization_pt_vq_k16"
OBJECTS_DIR_GT = "/home/xianghui_yang/data/NVF_data/"
OBJECTS_DIR = os.path.join(WORK_DIR, "mesh")
sys.stdout = open(os.path.join(WORK_DIR, "test_mesh.txt"), "w")
BASE_CLASSES = ["car","chair","plane", "table"]
NOVEL_CLASSES = ["speaker", "bench", "lamp", "watercraft"]
# BASE_CLASSES = ["car"]
# NOVEL_CLASSES = []
device = torch.device("cuda:0")
data_dir = '/home/xianghui_yang/data/ShapeNet/'
with open(os.path.join(data_dir, 'id2cat.json'), 'r') as f:
idx2cls = json.load(f)
cls2idx = {}
for idx in idx2cls.keys():
cls2idx[idx2cls[idx]]=idx
class TestDataset(data.Dataset):
def __init__(self, categories=["car"], object_dir="/home/xianghui_yang/data/ShapeNet/ShapeNetV1NDF_CUBE_OFFSET/", object_dir_gt="/home/xianghui_yang/data/ShapeNet/ShapeNetV1NDF_CUBE_OFFSET/"):
self.object_dir = object_dir
self.object_dir_gt = object_dir_gt
with open('/home/xianghui_yang/data/ShapeNet/id2cat.json', 'r') as f:
self.idx2cls = json.load(f)
self.cls2idx = {}
for idx in idx2cls.keys():
self.cls2idx[idx2cls[idx]]=idx
self.file_list = []
if len(categories)==1:
data_np = np.load("/home/xianghui_yang/data/NVF_data/split_02958343.npz")['test']
for fn in data_np:
self.file_list.append((fn, "car", "02958343"))
else:
for cat in categories:
idx = self.cls2idx[cat]
data_np = np.load("/home/xianghui_yang/data/NVF_data/split_generalization_%s.npz"%(idx))['test']
for fn in data_np:
self.file_list.append((fn, cat, idx))
print(categories)
print(len(self.file_list))
def __len__(self):
return len(self.file_list)
def __getitem__(self, idx):
fn, cat, idx = self.file_list[idx]
verts_pred, faces_pred, _ = load_obj(os.path.join(self.object_dir, idx, fn+".obj"), load_textures=False)
verts_gt, faces_gt, _ = load_obj(os.path.join(self.object_dir_gt, idx, fn+".obj"), load_textures=False)
voxel_path = os.path.join(self.object_dir_gt, idx, fn, 'voxelized_point_cloud_256res_10000points.npz')
input_points = torch.from_numpy(np.load(voxel_path)['point_cloud'].astype(np.float32))
return verts_pred, verts_gt, faces_pred.verts_idx, faces_gt.verts_idx, input_points, cat, fn
val_distance = {}
val_cd_median = {}
val_normal = {}
val_emd = {}
val_fscore_tau = {}
val_fscore_2tau = {}
val_fscore_3tau = {}
val_fscore_4tau = {}
for cat in BASE_CLASSES+NOVEL_CLASSES:
val_distance[cat] = AverageValueMeter()
val_cd_median[cat] = AverageValueMeter()
val_normal[cat] = AverageValueMeter()
val_emd[cat] = AverageValueMeter()
val_fscore_tau[cat] = AverageValueMeter()
val_fscore_2tau[cat] = AverageValueMeter()
val_fscore_3tau[cat] = AverageValueMeter()
val_fscore_4tau[cat] = AverageValueMeter()
dataset = TestDataset(categories=BASE_CLASSES+NOVEL_CLASSES, object_dir=OBJECTS_DIR, object_dir_gt=OBJECTS_DIR_GT)
dataloader = DataLoader(dataset, batch_size=1, shuffle=False, num_workers=4, drop_last=False)
for i, data in enumerate(dataloader):
verts_pred, verts_gt, faces_pred_idx, faces_gt_idx, input_points, cat, fn = data
if verts_pred.shape[1]==0 or faces_pred_idx.shape[1]==0:
print(cat[0], fn[0])
exit()
if REMOVE_OUTLIERS:
# print(cat[0], fn[0])
dist_vp, _, _, _= distChamfer(verts_pred.to(device=device), input_points.to(device=device))
dist_vp = dist_vp.squeeze(0).cpu()
# print(dist_vp.shape, faces_pred_idx.shape)
faces_pred_idx = faces_pred_idx[0]
dist_fp = dist_vp[faces_pred_idx]
# print(dist_fp.shape)
dist_fp = torch.max(dist_fp, dim=1)[0]
faces_pred_idx[dist_fp>REMOVE_OUTLIERS_THRESHOLD] = -1
faces_pred_idx = faces_pred_idx.unsqueeze(0)
# print(verts_pred.shape, faces_pred_idx.shape, verts_pred.device, faces_pred_idx.device)
verts_pred, verts_gt, faces_pred_idx, faces_gt_idx, cat, fn = \
verts_pred[0], verts_gt[0], faces_pred_idx[0], faces_gt_idx[0], cat[0], fn[0]
meshes = Meshes(verts=[verts_pred, verts_gt], faces=[faces_pred_idx, faces_gt_idx]).to(device=device)
pc, normals = sample_points_from_meshes(meshes, CD_NUM, return_normals=True)
pred_pc = pc[0].unsqueeze(0)
gt_pc = pc[1].unsqueeze(0)
pred_normals = normals[0].unsqueeze(0)
gt_normals = normals[1].unsqueeze(0)
# meshes = Meshes(verts=[verts_gt], faces=[faces_gt_idx]).to(device=device)
# gt_pc = sample_points_from_meshes(meshes, CD_NUM, return_normals=False)
# pred_pc = input_points.to(device=device)
# print(pred_pc.shape, gt_pc.shape)
d1, d2, _, _= distChamfer(pred_pc, gt_pc)
val_fscore = get_f_score(d1, d2, [1e-5, 2e-5, 2.5e-5, 1e-4])
val_pc_chamfer_loss = (torch.mean(d1) + torch.mean(d2))/2
val_pc_chamfer_loss, val_normal_loss = chamfer_distance(pred_pc, gt_pc, x_normals=pred_normals, y_normals=gt_normals)
val_pc_chamfer_loss /= 2
val_pc_chamfer_median = (torch.median(d1) + torch.median(d2)) /2
indices = np.random.choice(CD_NUM, size=EMD_NUM, replace=False)
val_emd_loss = earth_mover_distance(pred_pc[:, indices, :], gt_pc[:, indices, :], transpose=False)
val_distance[cat].update(val_pc_chamfer_loss.item())
val_cd_median[cat].update(val_pc_chamfer_median.item())
val_normal[cat].update(val_normal_loss.item())
val_emd[cat].update(val_emd_loss[0].item()/EMD_NUM)
val_fscore_tau[cat].update(val_fscore[0, 0].item())
val_fscore_2tau[cat].update(val_fscore[0, 1].item())
val_fscore_3tau[cat].update(val_fscore[0, 2].item())
val_fscore_4tau[cat].update(val_fscore[0, 3].item())
for cat in BASE_CLASSES+NOVEL_CLASSES:
print(cat)
print(val_distance[cat].count, val_normal[cat].count, val_emd[cat].count, val_fscore_tau[cat].count, val_fscore_2tau[cat].count)
def calculate_metrics(eval_classes, class_split="mean"):
if len(eval_classes)==0:
return
mean_chamfer = []
mean_median = []
mean_normal = []
mean_emd = []
mean_f_score_tau = []
mean_f_score_2tau = []
mean_f_score_3tau = []
mean_f_score_4tau = []
print("Class", "CD Mean", "CD Median", "Normal", "EMD", "F-score tau", "F-score 2tau")
for cat in eval_classes:
mean_chamfer.append(val_distance[cat].avg*1e4)
mean_median.append(val_cd_median[cat].avg*1e4)
mean_normal.append(val_normal[cat].avg)
mean_emd.append(val_emd[cat].avg*100)
mean_f_score_tau.append(val_fscore_tau[cat].avg)
mean_f_score_2tau.append(val_fscore_2tau[cat].avg)
mean_f_score_3tau.append(val_fscore_3tau[cat].avg)
mean_f_score_4tau.append(val_fscore_4tau[cat].avg)
print(cat, "%.4f %.4f %.4f %.4f %.4f %.4f %.4f %.4f"%(val_distance[cat].avg*1e4, val_cd_median[cat].avg*1e4, val_normal[cat].avg, val_emd[cat].avg*100,
val_fscore_tau[cat].avg, val_fscore_2tau[cat].avg, val_fscore_3tau[cat].avg, val_fscore_4tau[cat].avg))
print(class_split, 'cd mean: %.4f'%(np.mean(mean_chamfer)))
print(class_split, 'cd median: %.4f'%(np.mean(mean_median)))
print(class_split, 'normal: %.4f'%(np.mean(mean_normal)))
print(class_split, 'EMD: %.4f'%(np.mean(mean_emd)))
print(class_split, 'f-score 1e-5: %.4f'%(np.mean(mean_f_score_tau)))
print(class_split, 'f-score 2e-5: %.4f'%(np.mean(mean_f_score_2tau)))
print(class_split, 'f-score 2.5e-5: %.4f'%(np.mean(mean_f_score_3tau)))
print(class_split, 'f-score 1e-4: %.4f'%(np.mean(mean_f_score_4tau)))
return
print("=========Mean=========")
calculate_metrics(BASE_CLASSES+NOVEL_CLASSES, class_split="mean")
print("=========BASE=========")
calculate_metrics(BASE_CLASSES, class_split="base")
print("=========NOVEL=========")
calculate_metrics(NOVEL_CLASSES, class_split="novel")