smpl-core 0.9.0

Core functionality of smpl-rs
Documentation
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
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
use crate::{common::vertex_offsets::VertexOffsetsG, AppBackend};
use crate::{
    common::{
        betas::BetasG,
        expression::ExpressionG,
        outputs::SmplOutputG,
        pose::PoseG,
        smpl_model::{FaceModel, SmplModel},
        smpl_options::SmplOptions,
        types::{Gender, SmplType, UpAxis},
    },
    conversions::pose_remap::PoseRemap,
};
use burn::tensor::{backend::Backend, Float, Int, Tensor};
use gloss_geometry::csr::{VertexFaceCSR, VertexFaceCSRBurn};
use gloss_utils::bshare::ToBurn;
use gloss_utils::nshare::ToNalgebra;
use log::{info, warn};
use nalgebra as na;
use ndarray as nd;
use ndarray::prelude::*;
use ndarray_npy::NpzReader;
use smpl_utils::{
    array::Gather2D,
    io::FileLoader,
    numerical::{batch_rigid_transform_burn_fast, batch_rodrigues_burn_3},
};
use std::{
    any::Any,
    io::{Read, Seek},
};
pub const NUM_BODY_JOINTS: usize = 21;
pub const NUM_HAND_JOINTS: usize = 15;
pub const NUM_FACE_JOINTS: usize = 3;
pub const NUM_JOINTS: usize = NUM_BODY_JOINTS + 2 * NUM_HAND_JOINTS + NUM_FACE_JOINTS;
pub const NECK_IDX: usize = 12;
pub const NUM_VERTS: usize = 10475;
pub const NUM_VERTS_UV_MESH: usize = 11307;
pub const NUM_FACES: usize = 20908;
pub const FULL_SHAPE_SPACE_DIM: usize = 400;
pub const SHAPE_SPACE_DIM: usize = 300;
pub const EXPRESSION_SPACE_DIM: usize = 100;
pub const NUM_POSE_BLEND_SHAPES: usize = NUM_JOINTS * 9;
#[derive(Clone)]
pub struct SmplXGPUG<B: Backend> {
    pub device: B::Device,
    pub smpl_type: SmplType,
    pub gender: Gender,
    pub verts_template: Tensor<B, 2, Float>,
    pub faces: Tensor<B, 2, Int>,
    pub faces_uv_mesh: Tensor<B, 2, Int>,
    pub uv: Tensor<B, 2, Float>,
    pub shape_dirs: Tensor<B, 2, Float>,
    pub expression_dirs: Option<Tensor<B, 2, Float>>,
    pub pose_dirs: Option<Tensor<B, 2, Float>>,
    pub joint_regressor: Tensor<B, 2, Float>,
    pub parent_idx_per_joint_nd: nd::Array1<u32>,
    pub parent_idx_per_joint: Tensor<B, 1, Int>,
    pub lbs_weights: Tensor<B, 2, Float>,
    pub verts_ones: Tensor<B, 2, Float>,
    pub idx_vuv_2_vnouv: Tensor<B, 1, Int>,
    pub faces_na: na::DMatrix<u32>,
    pub faces_uv_mesh_na: na::DMatrix<u32>,
    pub uv_na: na::DMatrix<f32>,
    pub idx_vuv_2_vnouv_vec: Vec<usize>,
    pub lbs_weights_split: Tensor<B, 2>,
    pub lbs_weights_nd: nd::ArcArray2<f32>,
    pub lbs_weights_split_nd: nd::ArcArray2<f32>,
    pub vertex_face_csr: VertexFaceCSRBurn<B>,
    pub vertex_face_uv_csr: VertexFaceCSRBurn<B>,
    pub kinematic_tree_depth: usize,
}
impl<B: Backend> SmplXGPUG<B> {
    /// # Panics
    /// Will panic if the matrices don't match the expected sizes
    #[allow(clippy::too_many_arguments)]
    #[allow(clippy::too_many_lines)]
    pub fn new_from_matrices(
        gender: Gender,
        verts_template: &nd::Array2<f32>,
        faces: &nd::Array2<u32>,
        faces_uv_mesh: &nd::Array2<u32>,
        uv: &nd::Array2<f32>,
        shape_dirs: &nd::Array3<f32>,
        expression_dirs: Option<nd::Array3<f32>>,
        pose_dirs: Option<nd::Array3<f32>>,
        joint_regressor: &nd::Array2<f32>,
        parent_idx_per_joint: &nd::Array1<u32>,
        lbs_weights: nd::Array2<f32>,
        max_num_betas: usize,
        max_num_expression_components: usize,
    ) -> Self {
        let device = B::Device::default();
        let b_verts_template = verts_template.to_burn(&device);
        let b_faces = faces.to_burn(&device);
        let b_faces_uv_mesh = faces_uv_mesh.to_burn(&device);
        let b_uv = uv.to_burn(&device);
        let actual_num_betas = max_num_betas.min(shape_dirs.shape()[2]);
        let shape_dirs = shape_dirs
            .slice_axis(Axis(2), ndarray::Slice::from(0..actual_num_betas))
            .to_owned()
            .into_shape((NUM_VERTS * 3, actual_num_betas))
            .unwrap();
        let b_shape_dirs = shape_dirs.to_burn(&device);
        let b_expression_dirs = expression_dirs.map(|expression_dirs| {
            let actual_num_expression_components = max_num_expression_components.min(expression_dirs.shape()[2]);
            let expression_dirs = expression_dirs
                .slice_axis(nd::Axis(2), nd::Slice::from(0..actual_num_expression_components))
                .into_shape((NUM_VERTS * 3, actual_num_expression_components))
                .unwrap()
                .to_owned();
            expression_dirs.to_burn(&device)
        });
        let b_pose_dirs = pose_dirs.map(|pose_dirs| {
            let pose_dirs = pose_dirs.into_shape((NUM_VERTS * 3, NUM_JOINTS * 9)).unwrap();
            pose_dirs.to_burn(&device)
        });
        let b_joint_regressor = joint_regressor.to_burn(&device);
        let b_parent_idx_per_joint = parent_idx_per_joint.to_burn(&device).reshape([NUM_JOINTS + 1]);
        let b_lbs_weights = lbs_weights.to_burn(&device);
        #[allow(clippy::cast_possible_wrap)]
        let faces_uv_mesh_i32: nd::Array2<i32> = faces_uv_mesh.mapv(|x| x as i32);
        let ft: nd::ArcArray2<i32> = faces_uv_mesh_i32.into();
        let max_v_uv_idx = *ft.iter().max_by_key(|&x| x).unwrap();
        let max_v_uv_idx_usize = usize::try_from(max_v_uv_idx).unwrap_or_else(|_| panic!("Cannot cast max_v_uv_idx to usize"));
        let mut idx_vuv_2_vnouv = nd::ArcArray1::<i32>::zeros(max_v_uv_idx_usize + 1);
        for (fuv, fnouv) in ft.axis_iter(nd::Axis(0)).zip(faces.axis_iter(nd::Axis(0))) {
            let uv_0 = fuv[[0]];
            let uv_1 = fuv[[1]];
            let uv_2 = fuv[[2]];
            let nouv_0 = fnouv[[0]];
            let nouv_1 = fnouv[[1]];
            let nouv_2 = fnouv[[2]];
            idx_vuv_2_vnouv[usize::try_from(uv_0).unwrap_or_else(|_| panic!("Cannot cast uv_0 to usize"))] =
                i32::try_from(nouv_0).unwrap_or_else(|_| panic!("Cannot cast nouv_0 to i32"));
            idx_vuv_2_vnouv[usize::try_from(uv_1).unwrap_or_else(|_| panic!("Cannot cast uv_1 to usize"))] =
                i32::try_from(nouv_1).unwrap_or_else(|_| panic!("Cannot cast nouv_1 to i32"));
            idx_vuv_2_vnouv[usize::try_from(uv_2).unwrap_or_else(|_| panic!("Cannot cast uv_2 to usize"))] =
                i32::try_from(nouv_2).unwrap_or_else(|_| panic!("Cannot cast nouv_2 to i32"));
        }
        let idx_vuv_2_vnouv_vec: Vec<i32> = idx_vuv_2_vnouv.mapv(|x| x).into_raw_vec();
        let idx_vuv_2_vnouv_slice: &[i32] = &idx_vuv_2_vnouv_vec;
        let b_idx_vuv_2_vnouv = Tensor::<B, 1, Int>::from_ints(idx_vuv_2_vnouv_slice, &device);
        let idx_vuv_2_vnouv_vec: Vec<usize> = idx_vuv_2_vnouv
            .to_vec()
            .iter()
            .map(|&x| usize::try_from(x).unwrap_or_else(|_| panic!("Cannot cast negative value to usize")))
            .collect();
        let faces_na = faces.view().into_nalgebra().clone_owned().map(|x| x);
        let faces_uv_mesh_na = ft
            .view()
            .into_nalgebra()
            .clone_owned()
            .map(|x| u32::try_from(x).unwrap_or_else(|_| panic!("Cannot cast value to u32")));
        let uv_na = uv.view().into_nalgebra().clone_owned();
        let cols: Vec<usize> = (0..lbs_weights.ncols()).collect();
        let lbs_weights_split: nd::ArcArray2<f32> = lbs_weights.to_owned().gather(&idx_vuv_2_vnouv_vec, &cols).into();
        let b_lbs_weights_split =
            Tensor::<B, 1>::from_floats(lbs_weights_split.as_slice().unwrap(), &device).reshape([idx_vuv_2_vnouv_vec.len(), NUM_JOINTS + 1]);
        let verts_ones = Tensor::<B, 2>::ones([NUM_VERTS, 1], &device);
        let lbs_weights_nd: nd::ArcArray2<f32> = lbs_weights.into();
        let cols: Vec<usize> = (0..lbs_weights_nd.ncols()).collect();
        let lbs_weights_split_nd = lbs_weights_nd.to_owned().gather(&idx_vuv_2_vnouv_vec, &cols).into();
        let vertex_face_csr = VertexFaceCSR::from_faces(&faces_na.clone());
        let vertex_face_csr_burn = vertex_face_csr.to_burn(&device);
        let vertex_face_uv_csr = VertexFaceCSR::from_faces(&faces_uv_mesh_na.clone());
        let vertex_face_uv_csr_burn = vertex_face_uv_csr.to_burn(&device);
        let kinematic_tree_depth = smpl_utils::numerical::compute_tree_depth(parent_idx_per_joint);
        info!("Initialised burn on Backend: {:?}", B::name(&device));
        info!("Device: {:?}", &device);
        Self {
            smpl_type: SmplType::SmplX,
            gender,
            device,
            verts_template: b_verts_template,
            faces: b_faces,
            faces_uv_mesh: b_faces_uv_mesh,
            uv: b_uv,
            shape_dirs: b_shape_dirs,
            expression_dirs: b_expression_dirs,
            pose_dirs: b_pose_dirs,
            joint_regressor: b_joint_regressor,
            parent_idx_per_joint: b_parent_idx_per_joint,
            parent_idx_per_joint_nd: parent_idx_per_joint.clone(),
            lbs_weights: b_lbs_weights,
            verts_ones,
            idx_vuv_2_vnouv: b_idx_vuv_2_vnouv,
            faces_na,
            faces_uv_mesh_na,
            uv_na,
            idx_vuv_2_vnouv_vec,
            lbs_weights_split: b_lbs_weights_split,
            lbs_weights_nd,
            lbs_weights_split_nd,
            vertex_face_csr: vertex_face_csr_burn,
            vertex_face_uv_csr: vertex_face_uv_csr_burn,
            kinematic_tree_depth,
        }
    }
    /// # Panics
    /// Will panic if the path cannot be opened
    fn new_from_npz_reader<R: Read + Seek>(
        npz: &mut NpzReader<R>,
        gender: Gender,
        max_num_betas: usize,
        max_num_expression_components: usize,
    ) -> Self {
        let verts_template: nd::Array2<f32> = npz.by_name("v_template.npy").unwrap();
        let faces: nd::Array2<u32> = npz.by_name("f.npy").unwrap();
        let uv: nd::Array2<f32> = npz.by_name("vt.npy").unwrap();
        let full_shape_dirs: nd::Array3<f32> = npz.by_name("shapedirs.npy").unwrap();
        let (shape_dirs, expression_dirs) = if let Ok(expression_dirs) = npz.by_name("expressiondirs.npy") {
            (full_shape_dirs, Some(expression_dirs))
        } else {
            let num_available_betas = full_shape_dirs.shape()[2];
            let num_full_betas = 300;
            let num_betas_to_use = num_full_betas.min(max_num_betas).min(num_available_betas);
            let shape_dirs = full_shape_dirs.slice_axis(nd::Axis(2), nd::Slice::from(0..num_betas_to_use)).to_owned();
            let expression_dirs = if full_shape_dirs.shape()[2] > 300 {
                Some(
                    full_shape_dirs
                        .slice_axis(nd::Axis(2), nd::Slice::from(300..300 + max_num_expression_components.min(100)))
                        .to_owned(),
                )
            } else {
                None
            };
            (shape_dirs, expression_dirs)
        };
        let pose_dirs: Option<nd::Array3<f32>> = npz.by_name("posedirs.npy").ok();
        let joint_regressor: nd::Array2<f32> = npz.by_name("J_regressor.npy").unwrap();
        let parent_idx_per_joint: nd::Array2<i32> = npz.by_name("kintree_table.npy").unwrap();
        #[allow(clippy::cast_sign_loss)]
        let parent_idx_per_joint = parent_idx_per_joint.mapv(|x| x as u32);
        let parent_idx_per_joint = parent_idx_per_joint
            .slice_axis(nd::Axis(0), nd::Slice::from(0..1))
            .to_owned()
            .into_shape(NUM_JOINTS + 1)
            .unwrap();
        let lbs_weights: nd::Array2<f32> = npz.by_name("weights.npy").unwrap();
        let ft: nd::Array2<u32> = npz.by_name("ft.npy").unwrap();
        if pose_dirs.is_none() {
            warn!("No pose_dirs loaded from npz");
        }
        Self::new_from_matrices(
            gender,
            &verts_template,
            &faces,
            &ft,
            &uv,
            &shape_dirs,
            expression_dirs,
            pose_dirs,
            &joint_regressor,
            &parent_idx_per_joint,
            lbs_weights,
            max_num_betas,
            max_num_expression_components,
        )
    }
    #[cfg(not(target_arch = "wasm32"))]
    /// # Panics
    /// Will panic if the path cannot be opened
    pub fn new_from_npz(model_path: &str, gender: Gender, max_num_betas: usize, max_num_expression_components: usize) -> Self {
        let mut npz = NpzReader::new(std::fs::File::open(model_path).unwrap()).unwrap();
        Self::new_from_npz_reader(&mut npz, gender, max_num_betas, max_num_expression_components)
    }
    /// # Panics
    /// Will panic if the path cannot be opened
    /// Will panic if the translation and rotation do not cover the same number
    /// of timesteps
    #[allow(clippy::cast_possible_truncation)]
    pub async fn new_from_npz_async(model_path: &str, gender: Gender, max_num_betas: usize, max_num_expression_components: usize) -> Self {
        let reader = FileLoader::open(model_path).await;
        let mut npz = NpzReader::new(reader).unwrap();
        Self::new_from_npz_reader(&mut npz, gender, max_num_betas, max_num_expression_components)
    }
    /// # Panics
    /// Will panic if the path cannot be opened
    /// Will panic if the translation and rotation do not cover the same number
    /// of timesteps
    #[allow(clippy::cast_possible_truncation)]
    pub fn new_from_reader<R: Read + Seek>(reader: R, gender: Gender, max_num_betas: usize, max_num_expression_components: usize) -> Self {
        let mut npz = NpzReader::new(reader).unwrap();
        Self::new_from_npz_reader(&mut npz, gender, max_num_betas, max_num_expression_components)
    }
    /// # Panics
    /// Will panic if the path cannot be opened
    /// Will panic if the translation and rotation do not cover the same number
    /// of timesteps
    #[allow(clippy::cast_possible_truncation)]
    pub fn read_pose_dirs_from_reader<R: Read + Seek>(reader: R, device: &B::Device) -> Tensor<B, 2, Float> {
        let mut npz = NpzReader::new(reader).unwrap();
        let pose_dirs: Option<nd::Array3<f32>> = Some(npz.by_name("pose_dirs.npy").unwrap());
        let b_pose_dirs =
            pose_dirs.map(|pose_dirs| Tensor::<B, 1>::from_floats(pose_dirs.as_slice().unwrap(), device).reshape([NUM_VERTS * 3, NUM_JOINTS * 9]));
        b_pose_dirs.unwrap()
    }
}
impl<B: Backend> FaceModel<B> for SmplXGPUG<B> {
    #[allow(clippy::missing_panics_doc)]
    #[allow(non_snake_case)]
    #[allow(clippy::let_and_return)]
    fn expression2offsets(&self, expression: &ExpressionG<B>) -> Tensor<B, 2, Float> {
        let device = self.verts_template.device();
        let offsets = if let Some(ref expression_dirs) = self.expression_dirs {
            let input_nr_expression_coeffs = expression.expr_coeffs.dims()[0];
            let model_nr_expression_coeffs = expression_dirs.shape().dims[1];
            let nr_expression_coeffs = input_nr_expression_coeffs.min(model_nr_expression_coeffs);
            #[allow(clippy::single_range_in_vec_init)]
            let expr_sliced = expression.expr_coeffs.clone().slice([0..nr_expression_coeffs]);
            let expression_dirs_sliced = expression_dirs.clone().slice([0..expression_dirs.dims()[0], 0..nr_expression_coeffs]);
            let v_expr_offsets = expression_dirs_sliced.matmul(expr_sliced.reshape([-1, 1]));
            v_expr_offsets.reshape([NUM_VERTS, 3])
        } else {
            Tensor::<B, 2, Float>::zeros([NUM_VERTS, 3], &device)
        };
        offsets
    }
    fn get_face_model(&self) -> &dyn FaceModel<B> {
        self
    }
}
impl<B: Backend> SmplModel<B> for SmplXGPUG<B> {
    fn clone_dyn(&self) -> Box<dyn SmplModel<B>> {
        Box::new(self.clone())
    }
    fn as_any(&self) -> &dyn Any {
        self
    }
    fn smpl_type(&self) -> SmplType {
        self.smpl_type
    }
    fn gender(&self) -> Gender {
        self.gender
    }
    fn device(&self) -> B::Device {
        self.device.clone()
    }
    fn get_face_model(&self) -> &dyn FaceModel<B> {
        self
    }
    #[allow(clippy::missing_panics_doc)]
    #[allow(non_snake_case)]
    fn forward(
        &self,
        options: &SmplOptions,
        betas: &BetasG<B>,
        pose_raw: &PoseG<B>,
        expression: Option<&ExpressionG<B>>,
        vertex_offsets: Option<&VertexOffsetsG<B>>,
    ) -> SmplOutputG<B> {
        let mut verts_t_pose = self.betas2verts(betas);
        if let Some(expression) = expression {
            verts_t_pose = verts_t_pose + self.expression2offsets(expression);
        }
        if let Some(vertex_offsets) = vertex_offsets {
            verts_t_pose = verts_t_pose + vertex_offsets.strength * vertex_offsets.offsets.clone();
        }
        let pose_remap = PoseRemap::new(pose_raw.smpl_type, SmplType::SmplX);
        let pose = pose_remap.remap(pose_raw);
        let joints_t_pose = self.verts2joints(verts_t_pose.clone());
        if options.enable_pose_corrective {
            let verts_offset = self.compute_pose_correctives(&pose);
            verts_t_pose = verts_t_pose + verts_offset;
        }
        let (verts_posed_nd, joints_posed) = self.apply_pose(&verts_t_pose, &joints_t_pose, &self.lbs_weights, &pose);
        SmplOutputG {
            verts: verts_posed_nd,
            faces: self.faces.clone(),
            normals: None,
            uvs: None,
            joints: joints_posed,
        }
    }
    fn create_body_with_uv(&self, smpl_merged: &SmplOutputG<B>) -> SmplOutputG<B> {
        let cols_tensor = Tensor::<B, 1, Int>::from_ints([0, 1, 2], &self.device);
        let mapping_tensor = self.idx_split_2_merged();
        let v_burn_split = smpl_merged.verts.clone().select(0, mapping_tensor.clone());
        let v_burn_split = v_burn_split.select(1, cols_tensor.clone());
        let n_burn_split = smpl_merged
            .normals
            .as_ref()
            .map(|n| n.clone().select(0, mapping_tensor).select(1, cols_tensor));
        SmplOutputG {
            verts: v_burn_split,
            faces: self.faces_uv_mesh.clone(),
            normals: n_burn_split,
            uvs: Some(self.uv.clone()),
            joints: smpl_merged.joints.clone(),
        }
    }
    #[allow(clippy::missing_panics_doc)]
    #[allow(non_snake_case)]
    #[allow(clippy::let_and_return)]
    fn betas2verts(&self, betas: &BetasG<B>) -> Tensor<B, 2, Float> {
        let input_nr_betas = betas.betas.dims()[0];
        let model_nr_betas = self.shape_dirs.shape().dims[1];
        let nr_betas = input_nr_betas.min(model_nr_betas);
        #[allow(clippy::single_range_in_vec_init)]
        let betas_sliced = betas.betas.clone().slice([0..nr_betas]);
        let shape_dirs_sliced = self.shape_dirs.clone().slice([0..self.shape_dirs.dims()[0], 0..nr_betas]);
        let v_beta_offsets = shape_dirs_sliced.matmul(betas_sliced.reshape([-1, 1]));
        let v_beta_offsets_reshaped = v_beta_offsets.reshape([NUM_VERTS, 3]);
        let verts_t_pose = v_beta_offsets_reshaped.add(self.verts_template.clone());
        verts_t_pose
    }
    fn verts2joints(&self, verts_t_pose: Tensor<B, 2, Float>) -> Tensor<B, 2, Float> {
        self.joint_regressor.clone().matmul(verts_t_pose)
    }
    #[allow(clippy::missing_panics_doc)]
    fn compute_pose_correctives(&self, pose: &PoseG<B>) -> Tensor<B, 2, Float> {
        if let Some(pose_dirs) = &self.pose_dirs {
            let full_pose = &pose.joint_poses;
            assert!(
                full_pose.dims()[0] == NUM_JOINTS + 1,
                "The pose does not have the correct number of joints for this model. Maybe you need to add a PoseRemapper component?\n {:?} != {:?}",
                full_pose.dims()[0],
                NUM_JOINTS + 1
            );
            let b_pose_feature = self.compute_pose_feature(pose);
            let b_pose_feature = b_pose_feature.reshape([NUM_JOINTS * 9, 1]);
            let new_pose_dirs = pose_dirs.clone();
            let all_pose_offsets = new_pose_dirs.matmul(b_pose_feature);
            all_pose_offsets.reshape([NUM_VERTS, 3])
        } else {
            Tensor::<B, 2, Float>::zeros([NUM_VERTS, 3], &self.device)
        }
    }
    #[allow(clippy::missing_panics_doc)]
    fn compute_pose_feature(&self, pose: &PoseG<B>) -> Tensor<B, 1> {
        let full_pose = &pose.joint_poses;
        assert!(
            full_pose.dims()[0] == NUM_JOINTS + 1,
            "The pose does not have the correct number of joints for this model. Maybe you need to add a PoseRemapper component?\n {:?} != {:?}",
            full_pose.dims()[0],
            NUM_JOINTS + 1
        );
        let rot_mats = batch_rodrigues_burn_3(full_pose);
        let identity = Tensor::<B, 2>::eye(3, &self.device());
        (rot_mats.clone().slice([1..rot_mats.dims()[0], 0..3, 0..3]) - identity.unsqueeze_dim(0)).reshape([NUM_JOINTS * 9])
    }
    #[allow(clippy::missing_panics_doc)]
    #[allow(non_snake_case)]
    #[allow(clippy::cast_precision_loss)]
    #[allow(clippy::cast_sign_loss)]
    #[allow(clippy::too_many_lines)]
    #[allow(clippy::similar_names)]
    fn apply_pose(
        &self,
        verts_t_pose: &Tensor<B, 2, Float>,
        joints: &Tensor<B, 2, Float>,
        lbs_weights: &Tensor<B, 2, Float>,
        pose: &PoseG<B>,
    ) -> (Tensor<B, 2, Float>, Tensor<B, 2, Float>) {
        assert!(
            verts_t_pose.shape().dims[0] == lbs_weights.shape().dims[0],
            "Verts and LBS weights should match"
        );
        let full_pose = &pose.joint_poses;
        assert!(
            full_pose.dims()[0] == NUM_JOINTS + 1,
            "The pose does not have the correct number of joints for this model."
        );
        let full_pose: Tensor<B, 2> = pose.joint_poses.clone();
        let rot_mats_t = batch_rodrigues_burn_3(&full_pose);
        let (posed_joints, rel_transforms) = batch_rigid_transform_burn_fast(
            self.parent_idx_per_joint.clone(),
            &self.parent_idx_per_joint_nd,
            rot_mats_t,
            joints.clone(),
            self.kinematic_tree_depth,
        );
        let nr_verts = verts_t_pose.shape().dims[0];
        let A = rel_transforms.reshape([NUM_JOINTS + 1, 16]);
        let T = lbs_weights.clone().matmul(A).reshape([nr_verts, 4, 4]);
        let ones = Tensor::ones([nr_verts, 1], &self.device);
        let v_posed_h = Tensor::cat(vec![verts_t_pose.clone(), ones], 1).unsqueeze_dim(2);
        let verts_final_h = T.matmul(v_posed_h).squeeze(2);
        let verts_final = verts_final_h.slice([0..nr_verts, 0..3]);
        let trans_pose = pose.global_trans.clone().reshape([1, 3]);
        let mut verts_final = verts_final.clone() + trans_pose.clone();
        let mut posed_joints = posed_joints.clone() + trans_pose.clone();
        if pose.up_axis == UpAxis::Z {
            let vcol0: Tensor<B, 1> = verts_final.clone().slice([0..nr_verts, 0..1]).squeeze(1);
            let vcol1: Tensor<B, 1> = verts_final.clone().slice([0..nr_verts, 1..2]).squeeze(1);
            let vcol2: Tensor<B, 1> = verts_final.clone().slice([0..nr_verts, 2..3]).squeeze(1);
            let verts_new_col1 = vcol2;
            let verts_new_col2 = vcol1.mul_scalar(-1.0);
            verts_final = Tensor::stack::<2>(vec![vcol0, verts_new_col1, verts_new_col2], 1);
            let nr_joints = posed_joints.shape().dims[0];
            let jcol0: Tensor<B, 1> = posed_joints.clone().slice([0..nr_joints, 0..1]).squeeze(1);
            let jcol1: Tensor<B, 1> = posed_joints.clone().slice([0..nr_joints, 1..2]).squeeze(1);
            let jcol2: Tensor<B, 1> = posed_joints.clone().slice([0..nr_joints, 2..3]).squeeze(1);
            let joints_new_col1 = jcol2;
            let joints_new_col2 = jcol1.mul_scalar(-1.0);
            posed_joints = Tensor::stack::<2>(vec![jcol0, joints_new_col1, joints_new_col2], 1);
        }
        (verts_final, posed_joints)
    }
    fn faces(&self) -> &Tensor<B, 2, Int> {
        &self.faces
    }
    fn faces_uv(&self) -> &Tensor<B, 2, Int> {
        &self.faces_uv_mesh
    }
    fn uv(&self) -> &Tensor<B, 2, Float> {
        &self.uv
    }
    fn lbs_weights(&self) -> Tensor<B, 2, Float> {
        self.lbs_weights.clone()
    }
    fn lbs_weights_split(&self) -> Tensor<B, 2, Float> {
        self.lbs_weights_split.clone()
    }
    fn idx_split_2_merged(&self) -> Tensor<B, 1, Int> {
        self.idx_vuv_2_vnouv.clone()
    }
    fn idx_split_2_merged_vec(&self) -> &Vec<usize> {
        &self.idx_vuv_2_vnouv_vec
    }
    fn set_pose_dirs(&mut self, pose_dirs: Tensor<B, 2, Float>) {
        self.pose_dirs = Some(pose_dirs);
    }
    fn get_pose_dirs(&self) -> Tensor<B, 2, Float> {
        if let Some(pose_dirs_tensor) = self.pose_dirs.clone() {
            pose_dirs_tensor
        } else {
            panic!("pose_dirs is not available!");
        }
    }
    fn get_expression_dirs(&self) -> Option<Tensor<B, 2, Float>> {
        self.expression_dirs.clone()
    }
    fn vertex_face_csr(&self) -> Option<VertexFaceCSRBurn<B>> {
        Some(self.vertex_face_csr.clone())
    }
    fn vertex_face_uv_csr(&self) -> Option<VertexFaceCSRBurn<B>> {
        Some(self.vertex_face_uv_csr.clone())
    }
    fn kinematic_tree_depth(&self) -> usize {
        self.kinematic_tree_depth
    }
}
pub type SmplXGPU = SmplXGPUG<AppBackend>;