3DSR/test_dsr_depth.py at main · VitjanZ/3DSR · GitHub

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
import torch
from data_loader import MVTecDRAEMTestDataset
from torch.utils.data import DataLoader
import numpy as np
from dsr_model import SubspaceRestrictionModule, ImageReconstructionNetwork, AnomalyDetectionModule
from discrete_model import DiscreteLatentModel
from au_pro_util import calculate_au_pro
from sklearn.metrics import roc_auc_score, average_precision_score


#def test(obj_names):
def test(obj_names, mvtec_path, out_path, run_name_base):
    total_ap_pixel = []
    total_auroc_pixel = []
    total_ap = []
    total_auroc = []
    total_aupro = []

    dada_model_path = "./checkpoints/DADA_D.pckl"

    num_hiddens = 256
    num_residual_hiddens = 128
    num_residual_layers = 2
    embedding_dim = 256
    #num_embeddings = 1024
    num_embeddings = 2048
    commitment_cost = 0.25
    decay = 0.99

    model = DiscreteLatentModel(num_hiddens, num_residual_layers, num_residual_hiddens, num_embeddings,
                                embedding_dim,
                                commitment_cost, decay, in_channels=1, out_channels=1)

    model.cuda()
    model.load_state_dict(torch.load(dada_model_path, map_location='cuda:0'))
    model.eval()

    embedder_hi = model._vq_vae_bot
    embedder_lo = model._vq_vae_top

    for obj_name in obj_names:
        img_dim = 384
        run_name = run_name_base+obj_name+'_'


        sub_res_model_lo = SubspaceRestrictionModule(embedding_size=embedding_dim)
        sub_res_model_lo.load_state_dict(torch.load(out_path+run_name+"_recon_lo.pckl", map_location='cuda:0'))
        sub_res_model_hi = SubspaceRestrictionModule(embedding_size=embedding_dim)
        sub_res_model_hi.load_state_dict(torch.load(out_path+run_name+"_recon_hi.pckl", map_location='cuda:0'))
        sub_res_model_lo.cuda()
        sub_res_model_lo.eval()
        sub_res_model_hi.cuda()
        sub_res_model_hi.eval()


        # Define the anomaly detection module - UNet-based network
        #decoder_seg = AnomalyDetectionModule(in_channels=2, base_width=64)
        #decoder_seg = AnomalyDetectionModule(in_channels=2, base_width=128)
        decoder_seg = AnomalyDetectionModule(in_channels=2, base_width=32)
        decoder_seg.load_state_dict(torch.load(out_path+run_name+"_seg.pckl", map_location='cuda:0'))
        decoder_seg.cuda()
        decoder_seg.eval()


        # Image reconstruction network reconstructs the image from discrete features.
        # It is trained for a specific object
        model_decode = ImageReconstructionNetwork(embedding_dim * 2,
                   num_hiddens,
                   num_residual_layers,
                   num_residual_hiddens, out_channels=1)
        model_decode.load_state_dict(torch.load(out_path+run_name+"_decode.pckl", map_location='cuda:0'))
        model_decode.cuda()
        model_decode.eval()

        dataset = MVTecDRAEMTestDataset(mvtec_path + obj_name + "/test/*/xyz/",
                                        resize_shape=[img_dim, img_dim])
        dataloader = DataLoader(dataset, batch_size=1,
                                shuffle=False, num_workers=0)

        total_pixel_scores = np.zeros((img_dim * img_dim * len(dataset)))
        total_gt_pixel_scores = np.zeros((img_dim * img_dim * len(dataset)))
        total_pixel_scores_2d = np.zeros((len(dataset),img_dim, img_dim))
        total_gt_pixel_scores_2d = np.zeros((len(dataset),img_dim, img_dim))

        mask_cnt = 0

        total_gt = []
        total_score = []


        for i_batch, sample_batched in enumerate(dataloader):

            depth_image = sample_batched["image"].cuda()

            is_normal = sample_batched["has_anomaly"].detach().numpy()[0, 0]
            total_gt.append(is_normal)
            true_mask = sample_batched["mask"]
            true_mask_cv = true_mask.detach().numpy()[0, :, :, :].transpose((1, 2, 0))

            rgb_image = sample_batched["rgb_image"].cuda()


            in_image = torch.cat((depth_image, rgb_image), dim=1)
            in_image = depth_image
            _, _, recon_out, embeddings_lo, embeddings_hi = model(in_image)
            recon_image_general = recon_out

            _, recon_embeddings_hi, _ = sub_res_model_hi(embeddings_hi, embedder_hi)
            _, recon_embeddings_lo, _ = sub_res_model_lo(embeddings_lo, embedder_lo)

            # Reconstruct the image from the reconstructed features
            # with the object-specific image reconstruction module
            up_quantized_recon_t = model.upsample_t(recon_embeddings_lo)
            quant_join = torch.cat((up_quantized_recon_t, recon_embeddings_hi), dim=1)
            recon_image_recon = model_decode(quant_join)

            # Generate the anomaly segmentation map
            out_mask = decoder_seg(recon_image_recon.detach(), recon_image_general.detach())
            out_mask_sm = torch.softmax(out_mask, dim=1)
            out_mask_averaged = torch.nn.functional.avg_pool2d(out_mask_sm[:, 1:, :, :], 21, stride=1,
                                                               padding=21 // 2).cpu().detach().numpy()
            flat_out_mask = out_mask_averaged[0,0,:,:].flatten()
            image_score = np.max(out_mask_averaged)

            total_score.append(image_score)

            flat_true_mask = true_mask_cv.flatten()
            total_pixel_scores[mask_cnt * img_dim * img_dim:(mask_cnt + 1) * img_dim * img_dim] = flat_out_mask
            total_gt_pixel_scores[mask_cnt * img_dim * img_dim:(mask_cnt + 1) * img_dim * img_dim] = flat_true_mask
            total_pixel_scores_2d[mask_cnt] = out_mask_averaged[0,0,:,:]
            total_gt_pixel_scores_2d[mask_cnt] = true_mask_cv[:,:,0]
            mask_cnt += 1


        total_score = np.array(total_score)
        total_gt = np.array(total_gt)
        auroc = roc_auc_score(total_gt, total_score)
        ap = average_precision_score(total_gt, total_score)

        total_gt_pixel_scores = total_gt_pixel_scores.astype(np.uint8)
        total_gt_pixel_scores = total_gt_pixel_scores[:img_dim * img_dim * mask_cnt]
        total_pixel_scores = total_pixel_scores[:img_dim * img_dim * mask_cnt]
        auroc_pixel = roc_auc_score(total_gt_pixel_scores, total_pixel_scores)
        ap_pixel = average_precision_score(total_gt_pixel_scores, total_pixel_scores)
        aupro, _ = calculate_au_pro([total_gt_pixel_scores_2d[x] for x in range(total_gt_pixel_scores_2d.shape[0])], [total_pixel_scores_2d[x] for x in range(total_pixel_scores_2d.shape[0])])

        print("------------------")
        print(obj_name)
        print("AUC Image: " + str(auroc))
        print("AP Image: " + str(ap))
        print("AUC Pixel: " + str(auroc_pixel))
        print("AP Pixel: " + str(ap_pixel))
        print("AUPRO: " + str(aupro))

        total_aupro.append(aupro)
        total_auroc_pixel.append(auroc_pixel)
        total_auroc.append(auroc)
        total_ap.append(ap)
        total_ap_pixel.append(ap_pixel)
    print("--------MEAN---------------------------------------")
    print("AUC Image: " + str(np.mean(total_auroc)))
    print("AP Image: " + str(np.mean(total_ap)))
    print("AUC Pixel: " + str(np.mean(total_auroc_pixel)))
    print("AP Pixel: " + str(np.mean(total_ap_pixel)))
    print("AUPRO: " + str(np.mean(total_aupro)))

    print("AUC",*[np.round(x*100,2) for x in total_auroc],np.round(np.mean(total_auroc)*100,2))
    print("AUCp",*[np.round(x*100,2) for x in total_auroc_pixel],np.round(np.mean(total_auroc_pixel)*100,2))
    print("AUPRO",*[np.round(x*100,2) for x in total_aupro],np.round(np.mean(total_aupro)*100,2))
    print("AP",*[np.round(x*100,2) for x in total_ap],np.round(np.mean(total_ap)*100,2))


if __name__=="__main__":
    obj_names = ["bagel", "cable_gland", "carrot", "cookie", "dowel", "foam", "peach", "potato", "rope", "tire"]

    import argparse
    parser = argparse.ArgumentParser()
    parser.add_argument('--gpu_id', action='store', type=int, required=True)
    parser.add_argument('--data_path', action='store', type=str, required=True)
    parser.add_argument('--out_path', action='store', type=str, required=True)
    parser.add_argument('--run_name', action='store', type=str, required=True)

    args = parser.parse_args()
    with torch.cuda.device(args.gpu_id):
        test(obj_names,args.data_path, args.out_path, args.run_name)