File size: 15,009 Bytes
c7ae99a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
"""CMU ASF/AMC -> 272-dim representation (same layout as the HumanML3D-272 data used before).

Layout (as read by tvq272.py / gen_sample.py recovery):
  [0:2]    root xz velocity in the heading frame of the previous frame (frame 0: 0)
  [2:8]    heading change, 6D (first two rows): Rd_0 = H_0, Rd_t = H_t H_{t-1}^T
  [8:74]   joint positions, heading removed, root xz at origin (22x3)
  [74:140] joint velocities in the current heading frame (22x3)
  [140:272] parent-relative rotations 6D (first two rows); root = H_t G_root
Body sizes are kept as captured (the HumanML3D-272 data was not size-normalised either).
Every clip is also mirrored (left/right swapped, caption words left/right swapped).
30 fps. Floor: lowest joint of the clip at y = 0.

usage: python cmu_to272.py <all_asfamc dir> <cmu_files.csv> <out h272 dir>
"""
import os, sys, csv, re, json, math
import numpy as np

SCALE = (1.0 / 0.45) * 2.54 / 100.0          # CMU FAQ
J22 = ["pelvis", "l_hip", "r_hip", "spine1", "l_knee", "r_knee", "spine2", "l_ankle", "r_ankle", "spine3",
       "l_foot", "r_foot", "neck", "l_collar", "r_collar", "head", "l_shoulder", "r_shoulder", "l_elbow",
       "r_elbow", "l_wrist", "r_wrist"]
PAR = [-1, 0, 0, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 9, 9, 12, 13, 14, 16, 17, 18, 19]
# position of each of the 22 joints = end point of this CMU bone (root = root position)
POS = {0: "root", 1: "lhipjoint", 2: "rhipjoint", 3: "lowerback", 4: "lfemur", 5: "rfemur", 6: "upperback",
       7: "ltibia", 8: "rtibia", 9: "thorax", 10: "lfoot", 11: "rfoot", 12: "lowerneck", 15: "upperneck",
       16: "lclavicle", 17: "rclavicle", 18: "lhumerus", 19: "rhumerus", 20: "lradius", 21: "rradius"}
# collars (13, 14): 35% of the way from thorax end to clavicle end
# rotation of each joint = global orientation of the CMU bone that leaves it towards its main child
ROT = {0: "root", 1: "lfemur", 2: "rfemur", 3: "upperback", 4: "ltibia", 5: "rtibia", 6: "thorax",
       7: "lfoot", 8: "rfoot", 9: "lowerneck", 10: "ltoes", 11: "rtoes", 12: "upperneck", 13: "lclavicle",
       14: "rclavicle", 15: "head", 16: "lhumerus", 17: "rhumerus", 18: "lradius", 19: "rradius",
       20: "lhand", 21: "rhand"}
LR = [(1, 2), (4, 5), (7, 8), (10, 11), (13, 14), (16, 17), (18, 19), (20, 21)]


# ------------------------------------------------------------------ rotations
def rx(a):
    c, s = np.cos(a), np.sin(a); o, z = np.ones_like(a), np.zeros_like(a)
    return np.stack([np.stack([o, z, z], -1), np.stack([z, c, -s], -1), np.stack([z, s, c], -1)], -2)


def ry(a):
    c, s = np.cos(a), np.sin(a); o, z = np.ones_like(a), np.zeros_like(a)
    return np.stack([np.stack([c, z, s], -1), np.stack([z, o, z], -1), np.stack([-s, z, c], -1)], -2)


def rz(a):
    c, s = np.cos(a), np.sin(a); o, z = np.ones_like(a), np.zeros_like(a)
    return np.stack([np.stack([c, -s, z], -1), np.stack([s, c, z], -1), np.stack([z, z, o], -1)], -2)


def euler_xyz(x, y, z):           # rotate about X, then Y, then Z  ->  Rz Ry Rx (degrees in)
    return rz(np.radians(z)) @ ry(np.radians(y)) @ rx(np.radians(x))


# ------------------------------------------------------------------ ASF / AMC
def parse_asf(path):
    lines = [l.strip() for l in open(path, encoding="latin-1")]
    bones, hier, root_order, i = {}, {}, ["tx", "ty", "tz", "rx", "ry", "rz"], 0
    root_axis = "XYZ"
    while i < len(lines):
        l = lines[i]
        if l.startswith(":root"):
            i += 1
            while not lines[i].startswith(":"):
                t = lines[i].split()
                if t and t[0] == "order": root_order = [x.lower() for x in t[1:]]
                if t and t[0] == "axis": root_axis = t[1]
                i += 1
            continue
        if l.startswith(":bonedata"):
            i += 1
            while not lines[i].startswith(":"):
                if lines[i] == "begin":
                    b = dict(dof=[])
                    i += 1
                    while lines[i] != "end":
                        t = lines[i].split()
                        if t[0] == "name": b["name"] = t[1]
                        elif t[0] == "direction": b["dir"] = np.array(list(map(float, t[1:4])))
                        elif t[0] == "length": b["length"] = float(t[1]) * SCALE
                        elif t[0] == "axis":
                            assert t[4] == "XYZ", t
                            b["axis"] = np.array(list(map(float, t[1:4])))
                        elif t[0] == "dof": b["dof"] = [x.lower() for x in t[1:]]
                        i += 1
                    bones[b["name"]] = b
                i += 1
            continue
        if l.startswith(":hierarchy"):
            i += 1
            while i < len(lines) and lines[i] != "end":
                t = lines[i].split()
                if t and t[0] != "begin":
                    for ch in t[1:]: hier[ch] = t[0]
                i += 1
            continue
        i += 1
    assert root_axis == "XYZ"
    order = ["root"]
    k = 0
    while k < len(order):
        order += [c for c, p in hier.items() if p == order[k]]
        k += 1
    for b in bones.values():
        b["C"] = euler_xyz(*b["axis"])
    return bones, hier, order, root_order


def parse_amc(path):
    frames, cur = [], None
    for l in open(path, encoding="latin-1"):
        l = l.strip()
        if not l or l[0] in "#:": continue
        if l.isdigit():
            cur = {}; frames.append(cur); continue
        t = l.split()
        cur[t[0]] = [float(v) for v in t[1:]]
    return frames


def fk(bones, hier, order, root_order, frames):
    """vectorised over frames -> global rotation M[name] (T,3,3) and end point E[name] (T,3), meters"""
    T = len(frames)
    rv = np.array([f["root"] for f in frames])
    d = {k: rv[:, i] for i, k in enumerate(root_order)}
    M, E = {}, {}
    M["root"] = euler_xyz(d["rx"], d["ry"], d["rz"])            # root axis = 0 0 0 -> C = I
    E["root"] = np.stack([d["tx"], d["ty"], d["tz"]], -1) * SCALE
    for n in order[1:]:
        b, p = bones[n], hier[n]
        ang = {"rx": np.zeros(T), "ry": np.zeros(T), "rz": np.zeros(T)}
        if b["dof"]:
            v = np.array([f.get(n, [0.0] * len(b["dof"])) for f in frames])
            for k, dof in enumerate(b["dof"]):
                ang[dof] = v[:, k]
        L = euler_xyz(ang["rx"], ang["ry"], ang["rz"])
        M[n] = M[p] @ b["C"] @ L @ b["C"].T
        E[n] = E[p] + b["length"] * (M[n] @ b["dir"])
    return M, E


def to22(M, E):
    P = np.zeros((len(E["root"]), 22, 3))
    for j, n in POS.items():
        P[:, j] = E[n]
    P[:, 13] = E["thorax"] + 0.35 * (E["lclavicle"] - E["thorax"])
    P[:, 14] = E["thorax"] + 0.35 * (E["rclavicle"] - E["thorax"])
    G = np.stack([M[ROT[j]] for j in range(22)], 1)          # (T,22,3,3)
    return P, G


def mirror(P, G):
    S = np.diag([-1.0, 1.0, 1.0])
    P2, G2 = P * np.array([-1.0, 1.0, 1.0]), S @ G @ S
    for a, b in LR:
        P2[:, [a, b]] = P2[:, [b, a]]
        G2[:, [a, b]] = G2[:, [b, a]]
    return P2, G2


# ------------------------------------------------------------------ 272 features
def six(R):                         # first two ROWS
    return R[..., :2, :].reshape(*R.shape[:-2], 6)


def features(P, G):
    P = P.copy()
    P[..., 1] -= P[..., 1].min()                             # floor
    P[..., [0, 2]] -= P[0, 0, [0, 2]]                        # start at origin
    T = P.shape[0]
    f = G[:, 0] @ np.array([0.0, 0.0, 1.0])                  # body forward
    a = np.arctan2(f[:, 0], f[:, 2])
    H = np.transpose(ry(a), (0, 2, 1))                       # world -> heading frame
    r = P[:, 0].copy(); r[:, 1] = 0
    loc = np.einsum("tij,tkj->tki", H, P - r[:, None])
    dP = np.zeros_like(P); dP[1:] = P[1:] - P[:-1]; dP[0] = dP[1] if T > 1 else 0
    vel = np.einsum("tij,tkj->tki", H, dP)
    rv = np.zeros((T, 3))
    rv[1:] = np.einsum("tij,tj->ti", H[:-1], r[1:] - r[:-1])
    Rd = np.empty_like(H); Rd[0] = H[0]; Rd[1:] = H[1:] @ np.transpose(H[:-1], (0, 2, 1))
    R = np.empty_like(G)
    R[:, 0] = H @ G[:, 0]
    for j in range(1, 22):
        R[:, j] = np.transpose(G[:, PAR[j]], (0, 2, 1)) @ G[:, j]
    x = np.concatenate([rv[:, [0, 2]], six(Rd), loc.reshape(T, -1), vel.reshape(T, -1),
                        six(R).reshape(T, -1)], 1)
    assert x.shape[1] == 272
    return x.astype(np.float32), P, G


def recover(x):
    """numpy copy of gen_sample.py recovery -> world positions (T,22,3), global rotations"""
    def mat(s):
        a1, a2 = s[..., :3], s[..., 3:]
        b1 = a1 / np.linalg.norm(a1, axis=-1, keepdims=True)
        b2 = a2 - (b1 * a2).sum(-1, keepdims=True) * b1
        b2 /= np.linalg.norm(b2, axis=-1, keepdims=True)
        return np.stack([b1, b2, np.cross(b1, b2)], -2)
    T = x.shape[0]
    Rd = mat(x[:, 2:8])
    tot = [Rd[0]]
    for t in range(1, T):
        tot.append(Rd[t] @ tot[-1])
    inv = np.transpose(np.stack(tot), (0, 2, 1))
    Pl = x[:, 8:74].reshape(T, 22, 3)
    Pw = np.einsum("tij,tkj->tki", inv, Pl)
    v = np.zeros((T, 3)); v[:, 0], v[:, 2] = x[:, 0], x[:, 1]
    v[1:] = np.einsum("tij,tj->ti", inv[:-1], v[1:])
    tr = np.cumsum(v, 0)
    Pw[..., 0] += tr[:, None, 0]; Pw[..., 2] += tr[:, None, 2]
    Rl = mat(x[:, 140:272].reshape(T, 22, 6))
    Rl[:, 0] = inv @ Rl[:, 0]
    Gr = [None] * 22
    for j in range(22):
        Gr[j] = Rl[:, j] if PAR[j] < 0 else Gr[PAR[j]] @ Rl[:, j]
    return Pw, np.stack(Gr, 1)


# ------------------------------------------------------------------ captions
def clean_caption(s):
    s = re.sub(r"(?<=[a-z])(?=[A-Z])", " ", s.strip())       # CarefulWalk -> Careful Walk
    s = re.sub(r"\s+", " ", s).strip()
    return s


def mirror_caption(s):
    tmp = re.sub(r"\bleft\b", "\0L", s, flags=re.I)
    tmp = re.sub(r"\bright\b", "left", tmp, flags=re.I)
    tmp = tmp.replace("\0L", "right")
    tmp = re.sub(r"\bcounter[- ]?clockwise\b", "\0CC", tmp, flags=re.I)
    tmp = re.sub(r"\bclockwise\b", "counterclockwise", tmp, flags=re.I)
    return tmp.replace("\0CC", "clockwise")


# ------------------------------------------------------------------ main
def main():
    src, files_csv, out = sys.argv[1], sys.argv[2], sys.argv[3]
    meta = list(csv.DictReader(open(files_csv, encoding="utf-8-sig")))
    mdir = os.path.join(out, "motion_data", "cmu"); tdir = os.path.join(out, "texts", "texts")
    for d in (mdir, tdir, os.path.join(out, "split"), os.path.join(out, "mean_std")):
        os.makedirs(d, exist_ok=True)
    report, kept, asf_cache = [], [], {}
    for k, r in enumerate(meta):
        key, desc = r["file"], r["desc"].strip()
        rec = dict(file=key, desc=desc)
        if desc == "(一覧になし)":
            rec["skip"] = "not in CMU list (fps unknown)"; report.append(rec); continue
        fps = int(r["fps"]); step = fps // 30
        if fps % 30:
            rec["skip"] = "fps %d" % fps; report.append(rec); continue
        subj = key.split("_")[0]
        try:
            if subj not in asf_cache:
                asf_cache[subj] = parse_asf(os.path.join(src, "subjects", subj, subj + ".asf"))
            bones, hier, order, ro = asf_cache[subj]
            frames = parse_amc(os.path.join(src, "subjects", subj, key + ".amc"))[::step]
            if len(frames) < 16:
                rec["skip"] = "too short (%d frames)" % len(frames); report.append(rec); continue
            M, E = fk(bones, hier, order, ro, frames)
            P, G = to22(M, E)
        except Exception as e:                                   # noqa
            rec["skip"] = "error: %r" % e; report.append(rec); continue
        T = P.shape[0]
        sp = np.linalg.norm(np.diff(P, axis=0), axis=-1) * 30    # m/s
        rec.update(frames=T, sec=round(T / 30, 2), max_joint_speed=round(float(sp.max()), 2),
                   pelvis_h_median=round(float(np.median(P[:, 0, 1] - P[..., 1].min())), 3))
        flags = []
        if not np.isfinite(P).all() or not np.isfinite(G).all():
            flags.append("nan")
        if sp.max() > 20:
            flags.append("speed>20m/s")
        caption = "" if desc.lower() in ("", "clean") else clean_caption(desc)
        for tag, (PP, GG), cap in (("c", (P, G), caption), ("Mc", mirror(P, G), mirror_caption(caption))):
            x, Pn, Gn = features(PP, GG)
            Pr, Gr = recover(x)
            err = float(np.abs(Pr - Pn).max()) * 1000
            gerr = float(np.abs(Gr - Gn).max())
            if tag == "c":
                rec.update(recover_err_mm=round(err, 4), recover_rot_err=round(gerr, 6))
            if err > 1.0 or gerr > 1e-4:
                flags.append("recover %s %.3fmm %.1e" % (tag, err, gerr))
            if flags:
                continue
            np.save(os.path.join(mdir, tag + key + ".npy"), x)
            if cap:
                with open(os.path.join(tdir, tag + key + ".txt"), "w", encoding="utf-8") as f:
                    f.write("%s#%s#0.0#0.0\n" % (cap, cap.lower()))
        rec["flags"] = flags
        if not flags:
            kept.append(key)
        report.append(rec)
        if (k + 1) % 200 == 0:
            print("%d / %d" % (k + 1, len(meta)), flush=True)
    # split by original clip (mirror follows), 90/5/5
    rng = np.random.default_rng(0)
    ids = sorted(kept); rng.shuffle(ids)
    n = len(ids); nv = nt = max(1, round(n * 0.05))
    sp = {"val": ids[:nv], "test": ids[nv:nv + nt], "train": ids[nv + nt:]}
    for name, lst in sp.items():
        with open(os.path.join(out, "split", name + ".txt"), "w") as f:
            f.write("\n".join(t + i for i in lst for t in ("c", "Mc")) + "\n")
    # mean / std on train
    s1 = np.zeros(272); s2 = np.zeros(272); cnt = 0
    for i in sp["train"]:
        for t in ("c", "Mc"):
            x = np.load(os.path.join(mdir, t + i + ".npy")).astype(np.float64)
            s1 += x.sum(0); s2 += (x ** 2).sum(0); cnt += x.shape[0]
    mean = s1 / cnt; std = np.sqrt(np.maximum(s2 / cnt - mean ** 2, 0))
    np.save(os.path.join(out, "mean_std", "Mean.npy"), mean.astype(np.float32))
    np.save(os.path.join(out, "mean_std", "Std.npy"), np.maximum(std, 1e-3).astype(np.float32))
    json.dump(report, open(os.path.join(out, "convert_report.json"), "w", encoding="utf-8"), ensure_ascii=False, indent=0)
    flagged = [r for r in report if r.get("flags")]
    skipped = [r for r in report if r.get("skip")]
    sec = sum(r["sec"] for r in report if r.get("flags") == [])
    print("kept %d clips (%.2f h, x2 with mirror) | flagged %d | skipped %d | split train %d val %d test %d"
          % (len(kept), sec / 3600, len(flagged), len(skipped), len(sp["train"]), len(sp["val"]), len(sp["test"])))
    print("recover err max mm:", max(r.get("recover_err_mm", 0) for r in report))
    print("std floor hit dims:", int((std < 1e-3).sum()))
    for r in flagged[:20]:
        print("  FLAG", r["file"], r["flags"], r["desc"][:50])
    from collections import Counter
    print("skip reasons:", Counter(r["skip"].split(" (")[0].split(":")[0] for r in skipped))


if __name__ == "__main__":
    main()