File size: 5,141 Bytes
08764e9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
"""Visual comparison of reconstructed meshes vs ground truth for held-out cases.

Produces, per case, a PNG with GT (grey) and prediction (tooth=blue, canal=red)
overlaid, plus an error-colored view (vertices colored by distance to GT).

    python -m toothcanal.compare --config configs/default.yaml
    python -m toothcanal.compare --config configs/default.yaml --case 031-
"""
import os, glob, argparse
import numpy as np
import torch
from .utils import load_config, ensure_dir
from .splits import make_split
from .models import ImplicitNet
from .infer import reconstruct_instance
from .evaluate import gt_meshes


def _load_net(cfg, dev):
    ckpt = torch.load(os.path.join(cfg["paths"]["out_dir"], "stage2.pt"),
                      map_location=dev)
    n_lat = ckpt["model"]["latents.weight"].shape[0]

    class _N(ImplicitNet):
        def __init__(s):
            super().__init__(n_lat, cfg)
    net = _N().to(dev)
    net.load_state_dict(ckpt["model"]); net.eval()
    return net


def _nearest_dist(pred, gt, n=8000):
    from scipy.spatial import cKDTree
    import trimesh
    pp, _ = trimesh.sample.sample_surface(pred, n)
    gp, _ = trimesh.sample.sample_surface(gt, n)
    d, _ = cKDTree(gp).query(pp)
    return pp, d


def render_case(net, d, ids, cfg, dev, out_png, do_tto=True, max_teeth=None):
    import matplotlib
    matplotlib.use("Agg")
    import matplotlib.pyplot as plt
    from mpl_toolkits.mplot3d.art3d import Poly3DCollection

    fig = plt.figure(figsize=(16, 6))
    axO = fig.add_subplot(131, projection="3d"); axO.set_title("Overlay (grey=GT)")
    axP = fig.add_subplot(132, projection="3d"); axP.set_title("Prediction")
    axE = fig.add_subplot(133, projection="3d"); axE.set_title("Error (mm, vs GT)")

    allv = []
    errs = []
    teeth = ids if max_teeth is None else ids[:max_teeth]
    for iid in teeth:
        rec = reconstruct_instance(net, d, iid, cfg, dev, do_tto=do_tto)
        gt_t, gt_c = gt_meshes(d, iid, cfg)
        if rec is None or rec["tooth"] is None or gt_t is None:
            continue
        # place meshes in world by ROI origin so all teeth sit in the arch
        off = rec["roi"]["lo_world_mm"]
        pt = rec["tooth"].copy(); pt.vertices += off
        gt = gt_t.copy(); gt.vertices += off
        allv.append(pt.vertices)
        # GT grey
        axO.add_collection3d(Poly3DCollection(gt.vertices[gt.faces], facecolor=(.6,.6,.6),
                                              alpha=.25, linewidths=0))
        axO.add_collection3d(Poly3DCollection(pt.vertices[pt.faces], facecolor=(.2,.4,.8),
                                              alpha=.5, linewidths=0))
        axP.add_collection3d(Poly3DCollection(pt.vertices[pt.faces], facecolor=(.2,.4,.8),
                                              alpha=.7, linewidths=0))
        if rec["canal"] is not None:
            pc = rec["canal"].copy(); pc.vertices += off
            axP.add_collection3d(Poly3DCollection(pc.vertices[pc.faces], facecolor=(.85,.15,.15),
                                                  alpha=.95, linewidths=0))
            axO.add_collection3d(Poly3DCollection(pc.vertices[pc.faces], facecolor=(.85,.15,.15),
                                                  alpha=.8, linewidths=0))
        # error coloring
        pp, dd = _nearest_dist(pt, gt)
        errs.append(dd)
        axE.scatter(pp[:,0], pp[:,1], pp[:,2], c=dd, cmap="jet", s=1, vmin=0, vmax=1.0)

    if not allv:
        plt.close(fig); return False
    V = np.vstack(allv)
    for ax in (axO, axP, axE):
        ax.set_xlim(V[:,0].min(), V[:,0].max())
        ax.set_ylim(V[:,1].min(), V[:,1].max())
        ax.set_zlim(V[:,2].min(), V[:,2].max())
        ax.set_axis_off(); ax.view_init(elev=15, azim=-70)
    if errs:
        e = np.concatenate(errs)
        fig.suptitle(f"mean surface error = {e.mean():.3f} mm   "
                     f"(blue<{0.0:.1f}  red>{1.0:.1f} mm)", fontsize=12)
    fig.savefig(out_png, dpi=120, bbox_inches="tight"); plt.close(fig)
    return True


def main():
    ap = argparse.ArgumentParser()
    ap.add_argument("--config", default="configs/default.yaml")
    ap.add_argument("--case", default=None)
    ap.add_argument("--no_tto", action="store_true")
    ap.add_argument("--max_teeth", type=int, default=None,
                    help="limit teeth per case for a faster preview")
    args = ap.parse_args()
    cfg = load_config(args.config)
    dev = "cuda" if torch.cuda.is_available() else "cpu"
    net = _load_net(cfg, dev)

    _, test = make_split(cfg["paths"]["proc_dir"], cfg)
    cases = [args.case] if args.case else test
    out_dir = ensure_dir(os.path.join(cfg["paths"]["out_dir"], "compare"))
    for cid in cases:
        d = dict(np.load(os.path.join(cfg["paths"]["proc_dir"], f"{cid}.npz")))
        ids = [int(v) for v in np.unique(d["inst"]) if v > 0]
        out = os.path.join(out_dir, f"{cid}_compare.png")
        ok = render_case(net, d, ids, cfg, dev, out, do_tto=not args.no_tto,
                         max_teeth=args.max_teeth)
        print(f"[compare] {cid}: {'wrote '+out if ok else 'no meshes'}")


if __name__ == "__main__":
    main()