File size: 34,019 Bytes
2568013
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
from dataclasses import dataclass
from pathlib import Path
import gc
import random
from typing import Literal, Optional, Protocol, runtime_checkable, Any

import moviepy.editor as mpy
import torch
import torchvision
import wandb
from einops import pack, rearrange, repeat
from jaxtyping import Float
from lightning.pytorch import LightningModule
from lightning.pytorch.loggers.wandb import WandbLogger
from lightning.pytorch.utilities import rank_zero_only
from tabulate import tabulate
from torch import Tensor, nn, optim
import torch.nn.functional as F

from loss.loss_lpips import LossLpips
from loss.loss_mse import LossMse
from model.encoder.vggt.utils.pose_enc import pose_encoding_to_extri_intri

from ..loss.loss_distill import DistillLoss
from src.utils.render import generate_path
from src.utils.point import get_normal_map

from ..loss.loss_huber import HuberLoss, extri_intri_to_pose_encoding

# from model.types import Gaussians

from ..dataset.data_module import get_data_shim
from ..dataset.types import BatchedExample
from ..evaluation.metrics import compute_lpips, compute_psnr, compute_ssim, abs_relative_difference, delta1_acc
from ..global_cfg import get_cfg
from ..loss import Loss
from ..loss.loss_point import Regr3D
from ..loss.loss_ssim import ssim
from ..misc.benchmarker import Benchmarker
from ..misc.cam_utils import update_pose, get_pnp_pose, rotation_6d_to_matrix
from ..misc.image_io import prep_image, save_image, save_video
from ..misc.LocalLogger import LOG_PATH, LocalLogger
from ..misc.nn_module_tools import convert_to_buffer
from ..misc.step_tracker import StepTracker
from ..misc.utils import inverse_normalize, vis_depth_map, confidence_map, get_overlap_tag
from ..visualization.annotation import add_label
from ..visualization.camera_trajectory.interpolation import (
    interpolate_extrinsics,
    interpolate_intrinsics,
)
from ..visualization.camera_trajectory.wobble import (
    generate_wobble,
    generate_wobble_transformation,
)
from ..visualization.color_map import apply_color_map_to_image
from ..visualization.layout import add_border, hcat, vcat
# from ..visualization.validation_in_3d import render_cameras, render_projections
from .decoder.decoder import Decoder, DepthRenderingMode
from .encoder import Encoder
from .encoder.visualization.encoder_visualizer import EncoderVisualizer
from .ply_export import export_ply

@dataclass
class OptimizerCfg:
    lr: float
    warm_up_steps: int
    backbone_lr_multiplier: float


@dataclass
class TestCfg:
    output_path: Path
    align_pose: bool
    pose_align_steps: int
    rot_opt_lr: float
    trans_opt_lr: float
    compute_scores: bool
    save_image: bool
    save_video: bool
    save_compare: bool
    generate_video: bool
    mode: Literal["inference", "evaluation"]
    image_folder: str


@dataclass
class TrainCfg:
    output_path: Path
    depth_mode: DepthRenderingMode | None
    extended_visualization: bool
    print_log_every_n_steps: int
    distiller: str
    distill_max_steps: int
    pose_loss_alpha: float = 1.0
    pose_loss_delta: float = 1.0
    cxt_depth_weight: float = 0.01
    weight_pose: float = 1.0
    weight_depth: float = 1.0
    weight_normal: float = 1.0
    render_ba: bool = False
    render_ba_after_step: int = 0


@runtime_checkable
class TrajectoryFn(Protocol):
    def __call__(
        self,
        t: Float[Tensor, " t"],
    ) -> tuple[
        Float[Tensor, "batch view 4 4"],  # extrinsics
        Float[Tensor, "batch view 3 3"],  # intrinsics
    ]:
        pass


class ModelWrapper(LightningModule):
    logger: Optional[WandbLogger]
    model: nn.Module
    losses: nn.ModuleList
    optimizer_cfg: OptimizerCfg
    test_cfg: TestCfg
    train_cfg: TrainCfg
    step_tracker: StepTracker | None

    def __init__(
        self,
        optimizer_cfg: OptimizerCfg,
        test_cfg: TestCfg,
        train_cfg: TrainCfg,
        model: nn.Module,
        losses: list[Loss],
        step_tracker: StepTracker | None
    ) -> None:
        super().__init__()
        self.optimizer_cfg = optimizer_cfg
        self.test_cfg = test_cfg
        self.train_cfg = train_cfg
        self.step_tracker = step_tracker
        
        # Set up the model.
        self.encoder_visualizer = None
        self.model = model
        self.data_shim = get_data_shim(self.model.encoder)
        self.losses = nn.ModuleList(losses)
        
        if self.model.encoder.pred_pose:
            self.loss_pose = HuberLoss(alpha=self.train_cfg.pose_loss_alpha, delta=self.train_cfg.pose_loss_delta)
        
        if self.model.encoder.distill:
            self.loss_distill = DistillLoss(
                delta=self.train_cfg.pose_loss_delta,
                weight_pose=self.train_cfg.weight_pose,
                weight_depth=self.train_cfg.weight_depth,
                weight_normal=self.train_cfg.weight_normal
            )

        # This is used for testing.
        self.benchmarker = Benchmarker()
        
    def on_train_epoch_start(self) -> None:
        # our custom dataset and sampler has to have epoch set by calling set_epoch
        if hasattr(self.trainer.datamodule.train_loader.dataset, "set_epoch"):
            self.trainer.datamodule.train_loader.dataset.set_epoch(self.current_epoch)
        if hasattr(self.trainer.datamodule.train_loader.sampler, "set_epoch"):
            self.trainer.datamodule.train_loader.sampler.set_epoch(self.current_epoch)

    def on_validation_epoch_start(self) -> None:
        print(f"Validation epoch start on rank {self.trainer.global_rank}")
        # our custom dataset and sampler has to have epoch set by calling set_epoch
        if hasattr(self.trainer.datamodule.val_loader.dataset, "set_epoch"):
            self.trainer.datamodule.val_loader.dataset.set_epoch(self.current_epoch)
        if hasattr(self.trainer.datamodule.val_loader.sampler, "set_epoch"):
            self.trainer.datamodule.val_loader.sampler.set_epoch(self.current_epoch)
        
    def training_step(self, batch, batch_idx):
        # combine batch from different dataloaders
        # torch.cuda.empty_cache()
        if isinstance(batch, list):
            batch_combined = None
            for batch_per_dl in batch:
                if batch_combined is None:
                    batch_combined = batch_per_dl
                else:
                    for k in batch_combined.keys():
                        if isinstance(batch_combined[k], list):
                            batch_combined[k] += batch_per_dl[k]
                        elif isinstance(batch_combined[k], dict):
                            for kk in batch_combined[k].keys():
                                batch_combined[k][kk] = torch.cat([batch_combined[k][kk], batch_per_dl[k][kk]], dim=0)
                        else:
                            raise NotImplementedError
            batch = batch_combined
        
        batch: BatchedExample = self.data_shim(batch)
        b, v, c, h, w = batch["context"]["image"].shape
        context_image = (batch["context"]["image"] + 1) / 2
        
        # Run the model.
        visualization_dump = None

        encoder_output, output = self.model(context_image, self.global_step, visualization_dump=visualization_dump)
        gaussians, pred_pose_enc_list, depth_dict = encoder_output.gaussians, encoder_output.pred_pose_enc_list, encoder_output.depth_dict
        pred_context_pose = encoder_output.pred_context_pose
        infos = encoder_output.infos
        distill_infos = encoder_output.distill_infos
        
        num_context_views = pred_context_pose['extrinsic'].shape[1]

        using_index = torch.arange(num_context_views, device=gaussians.means.device)
        batch["using_index"] = using_index
        
        target_gt = (batch["context"]["image"] + 1) / 2
        scene_scale = infos["scene_scale"]
        self.log("train/scene_scale", infos["scene_scale"])
        self.log("train/voxelize_ratio", infos["voxelize_ratio"])

        # Compute metrics.
        psnr_probabilistic = compute_psnr(
            rearrange(target_gt, "b v c h w -> (b v) c h w"),
            rearrange(output.color, "b v c h w -> (b v) c h w"),
        )
        self.log("train/psnr_probabilistic", psnr_probabilistic.mean())

        consis_absrel = abs_relative_difference(
            rearrange(output.depth, "b v h w -> (b v) h w"),
            rearrange(depth_dict['depth'].squeeze(-1), "b v h w -> (b v) h w"),
            rearrange(distill_infos['conf_mask'], "b v h w -> (b v) h w"),
        )
        self.log("train/consis_absrel", consis_absrel.mean())

        consis_delta1 = delta1_acc(
            rearrange(output.depth, "b v h w -> (b v) h w"),
            rearrange(depth_dict['depth'].squeeze(-1), "b v h w -> (b v) h w"),
            rearrange(distill_infos['conf_mask'], "b v h w -> (b v) h w"),
        )
        self.log("train/consis_delta1", consis_delta1.mean())
        
        # Compute and log loss.
        total_loss = 0

        depth_dict['distill_infos'] = distill_infos
        with torch.amp.autocast('cuda', enabled=False):
            for loss_fn in self.losses:
                loss = loss_fn.forward(output, batch, gaussians, depth_dict, self.global_step)
                self.log(f"loss/{loss_fn.name}", loss)
                total_loss = total_loss + loss

            if depth_dict is not None and "depth" in get_cfg()["loss"].keys() and self.train_cfg.cxt_depth_weight > 0:
                depth_loss_idx = list(get_cfg()["loss"].keys()).index("depth")
                depth_loss_fn = self.losses[depth_loss_idx].ctx_depth_loss
                loss_depth = depth_loss_fn(depth_dict["depth_map"], depth_dict["depth_conf"], batch, cxt_depth_weight=self.train_cfg.cxt_depth_weight)
                self.log("loss/ctx_depth", loss_depth)
                total_loss = total_loss + loss_depth

            if distill_infos is not None:
                # distill ctx pred_pose & depth & normal
                loss_distill_list = self.loss_distill(distill_infos, pred_pose_enc_list, output, batch)
                self.log("loss/distill", loss_distill_list['loss_distill'])
                self.log("loss/distill_pose", loss_distill_list['loss_pose'])
                self.log("loss/distill_depth", loss_distill_list['loss_depth'])
                self.log("loss/distill_normal", loss_distill_list['loss_normal'])
                total_loss = total_loss + loss_distill_list['loss_distill']
        
        self.log("loss/total", total_loss)
        print(f"total_loss: {total_loss}")

        # Skip batch if loss is too high after certain step
        SKIP_AFTER_STEP = 1000  
        LOSS_THRESHOLD = 0.2
        if self.global_step > SKIP_AFTER_STEP and total_loss > LOSS_THRESHOLD:
            print(f"Skipping batch with high loss ({total_loss:.6f}) at step {self.global_step} on Rank {self.global_rank}")
            # set to a really small number
            return total_loss * 1e-10

        if (
            self.global_rank == 0
            and self.global_step % self.train_cfg.print_log_every_n_steps == 0
        ):
            print(
                f"train step {self.global_step}; "
                f"scene = {[x[:20] for x in batch['scene']]}; "
                f"context = {batch['context']['index'].tolist()}; "
                f"loss = {total_loss:.6f}; "
            )
            
        self.log("info/global_step", self.global_step)  # hack for ckpt monitor
        
        # Tell the data loader processes about the current step.
        if self.step_tracker is not None:
            self.step_tracker.set_step(self.global_step)
        
        del batch
        if self.global_step % 50 == 0:
            gc.collect()
            torch.cuda.empty_cache()

        return total_loss
    
    def on_after_backward(self):
        total_norm = 0.0
        counter = 0
        for p in self.parameters():
            if p.grad is not None:
                param_norm = p.grad.detach().data.norm(2)
                total_norm += param_norm.item() ** 2
                counter += 1
        total_norm = (total_norm / counter) ** 0.5
        self.log("loss/grad_norm", total_norm)
        
    def test_step(self, batch, batch_idx):
        batch: BatchedExample = self.data_shim(batch)
        b, v, _, h, w = batch["target"]["image"].shape
        assert b == 1
        if batch_idx % 100 == 0:
            print(f"Test step {batch_idx:0>6}.")
        
        # Render Gaussians.
        with self.benchmarker.time("encoder"):
            gaussians = self.model.encoder(
                (batch["context"]["image"]+1)/2,
                self.global_step,
            )[0]
        # export_ply(gaussians.means[0], gaussians.scales[0], gaussians.rotations[0], gaussians.harmonics[0], gaussians.opacities[0], Path("gaussians.ply"))
        # align the target pose
        if self.test_cfg.align_pose:
            output = self.test_step_align(batch, gaussians)
        else:
            with self.benchmarker.time("decoder", num_calls=v):
                output = self.model.decoder.forward(
                    gaussians,
                    batch["target"]["extrinsics"],
                    batch["target"]["intrinsics"],
                    batch["target"]["near"],
                    batch["target"]["far"],
                    (h, w),
                )
        
        # compute scores
        if self.test_cfg.compute_scores:
            overlap = batch["context"]["overlap"][0]
            overlap_tag = get_overlap_tag(overlap)

            rgb_pred = output.color[0]
            rgb_gt = batch["target"]["image"][0]
            all_metrics = {
                f"lpips_ours": compute_lpips(rgb_gt, rgb_pred).mean(),
                f"ssim_ours": compute_ssim(rgb_gt, rgb_pred).mean(),
                f"psnr_ours": compute_psnr(rgb_gt, rgb_pred).mean(),
            }
            methods = ['ours']

            self.log_dict(all_metrics)
            self.print_preview_metrics(all_metrics, methods, overlap_tag=overlap_tag)
        
        # Save images.
        (scene,) = batch["scene"]
        name = get_cfg()["wandb"]["name"]
        path = self.test_cfg.output_path / name
        if self.test_cfg.save_image:
            for index, color in zip(batch["target"]["index"][0], output.color[0]):
                save_image(color, path / scene / f"color/{index:0>6}.png")

        if self.test_cfg.save_video:
            frame_str = "_".join([str(x.item()) for x in batch["context"]["index"][0]])
            save_video(
                [a for a in output.color[0]],
                path / "video" / f"{scene}_frame_{frame_str}.mp4",
            )

        if self.test_cfg.save_compare:
            # Construct comparison image.
            context_img = inverse_normalize(batch["context"]["image"][0])
            comparison = hcat(
                add_label(vcat(*context_img), "Context"),
                add_label(vcat(*rgb_gt), "Target (Ground Truth)"),
                add_label(vcat(*rgb_pred), "Target (Prediction)"),
            )
            save_image(comparison, path / f"{scene}.png")
                
    def test_step_align(self, batch, gaussians):
        self.model.encoder.eval()
        # freeze all parameters
        for param in self.model.encoder.parameters():
            param.requires_grad = False

        b, v, _, h, w = batch["target"]["image"].shape
        output_c2ws = batch["target"]["extrinsics"]
        with torch.set_grad_enabled(True):
            cam_rot_delta = nn.Parameter(torch.zeros([b, v, 6], requires_grad=True, device=output_c2ws.device))
            cam_trans_delta = nn.Parameter(torch.zeros([b, v, 3], requires_grad=True, device=output_c2ws.device))
            opt_params = []
            self.register_buffer("identity", torch.tensor([1.0, 0.0, 0.0, 0.0, 1.0, 0.0]).to(output_c2ws))
            opt_params.append(
                {
                    "params": [cam_rot_delta],
                    "lr": 0.005,
                }
            )
            opt_params.append(
                {
                    "params": [cam_trans_delta],
                    "lr": 0.005,
                }
            )
            pose_optimizer = torch.optim.Adam(opt_params)
            extrinsics = output_c2ws.clone()
            with self.benchmarker.time("optimize"):
                for i in range(self.test_cfg.pose_align_steps):
                    pose_optimizer.zero_grad()
                    dx, drot = cam_trans_delta, cam_rot_delta
                    rot = rotation_6d_to_matrix(
                        drot + self.identity.expand(b, v, -1)
                    )  # (..., 3, 3)

                    transform = torch.eye(4, device=extrinsics.device).repeat((b, v, 1, 1))
                    transform[..., :3, :3] = rot
                    transform[..., :3, 3] = dx

                    new_extrinsics = torch.matmul(extrinsics, transform)
                    output = self.model.decoder.forward(
                        gaussians,
                        new_extrinsics,
                        batch["target"]["intrinsics"],
                        batch["target"]["near"],
                        batch["target"]["far"],
                        (h, w),
                        # cam_rot_delta=cam_rot_delta,
                        # cam_trans_delta=cam_trans_delta,
                    )

                    # Compute and log loss.
                    total_loss = 0
                    for loss_fn in self.losses:
                        loss = loss_fn.forward(output, batch, gaussians, self.global_step)
                        total_loss = total_loss + loss

                    total_loss.backward()
                    pose_optimizer.step()
                    
        # Render Gaussians.
        output = self.model.decoder.forward(
            gaussians,
            new_extrinsics,
            batch["target"]["intrinsics"],
            batch["target"]["near"],
            batch["target"]["far"],
            (h, w),
        )

        return output

    def on_test_end(self) -> None:
        name = get_cfg()["wandb"]["name"]
        self.benchmarker.dump(self.test_cfg.output_path / name / "benchmark.json")
        self.benchmarker.dump_memory(
            self.test_cfg.output_path / name / "peak_memory.json"
        )
        self.benchmarker.summarize()

    @rank_zero_only
    def validation_step(self, batch, batch_idx, dataloader_idx=0):        
        batch: BatchedExample = self.data_shim(batch)

        if self.global_rank == 0:
            print(
                f"validation step {self.global_step}; "
                f"scene = {batch['scene']}; "
                f"context = {batch['context']['index'].tolist()}"
            )

        # Render Gaussians.
        b, v, _, h, w = batch["context"]["image"].shape
        assert b == 1
        visualization_dump = {}

        encoder_output, output = self.model(batch["context"]["image"], self.global_step, visualization_dump=visualization_dump)
        gaussians, pred_pose_enc_list, depth_dict = encoder_output.gaussians, encoder_output.pred_pose_enc_list, encoder_output.depth_dict
        pred_context_pose, distill_infos = encoder_output.pred_context_pose, encoder_output.distill_infos
        infos = encoder_output.infos

        GS_num = infos['voxelize_ratio'] * (h*w*v)
        self.log("val/GS_num", GS_num)
        
        num_context_views = pred_context_pose['extrinsic'].shape[1]
        num_target_views = batch["target"]["extrinsics"].shape[1]
        rgb_pred = output.color[0].float()
        depth_pred = vis_depth_map(output.depth[0])

        # direct depth from gaussian means (used for visualization only)
        gaussian_means = visualization_dump["depth"][0].squeeze()
        if gaussian_means.shape[-1] == 3:
            gaussian_means = gaussian_means.mean(dim=-1)

        # Compute validation metrics.
        rgb_gt = (batch["context"]["image"][0].float() + 1) / 2
        psnr = compute_psnr(rgb_gt, rgb_pred).mean()
        self.log(f"val/psnr", psnr)
        lpips = compute_lpips(rgb_gt, rgb_pred).mean()
        self.log(f"val/lpips", lpips)
        ssim = compute_ssim(rgb_gt, rgb_pred).mean()
        self.log(f"val/ssim", ssim)

        # depth metrics
        consis_absrel = abs_relative_difference(
            rearrange(output.depth, "b v h w -> (b v) h w"),
            rearrange(depth_dict['depth'].squeeze(-1), "b v h w -> (b v) h w"),
        )
        self.log("val/consis_absrel", consis_absrel.mean())
        
        consis_delta1 = delta1_acc(
            rearrange(output.depth, "b v h w -> (b v) h w"),
            rearrange(depth_dict['depth'].squeeze(-1), "b v h w -> (b v) h w"),
            valid_mask=rearrange(torch.ones_like(output.depth, device=output.depth.device, dtype=torch.bool), "b v h w -> (b v) h w"),
        )
        self.log("val/consis_delta1", consis_delta1.mean())

        diff_map = torch.abs(output.depth - depth_dict['depth'].squeeze(-1))
        self.log("val/consis_mse", diff_map[distill_infos['conf_mask']].mean())

        # Construct comparison image.
        context_img = inverse_normalize(batch["context"]["image"][0])
        # context_img_depth = vis_depth_map(gaussian_means)
        context = []
        for i in range(context_img.shape[0]):
            context.append(context_img[i])
            # context.append(context_img_depth[i])
        
        colored_diff_map = vis_depth_map(diff_map[0], near=torch.tensor(1e-4, device=diff_map.device), far=torch.tensor(1.0, device=diff_map.device))
        model_depth_pred = depth_dict["depth"].squeeze(-1)[0]
        model_depth_pred = vis_depth_map(model_depth_pred)
        
        render_normal = (get_normal_map(output.depth.flatten(0, 1), batch["context"]["intrinsics"].flatten(0, 1)).permute(0, 3, 1, 2) + 1) / 2.
        pred_normal = (get_normal_map(depth_dict['depth'].flatten(0, 1).squeeze(-1), batch["context"]["intrinsics"].flatten(0, 1)).permute(0, 3, 1, 2) + 1) / 2.

        comparison = hcat(
            add_label(vcat(*context), "Context"),
            add_label(vcat(*rgb_gt), "Target (Ground Truth)"),
            add_label(vcat(*rgb_pred), "Target (Prediction)"),
            add_label(vcat(*depth_pred), "Depth (Prediction)"),
            add_label(vcat(*model_depth_pred), "Depth (VGGT Prediction)"),
            add_label(vcat(*render_normal), "Normal (Prediction)"),
            add_label(vcat(*pred_normal), "Normal (VGGT Prediction)"),
            add_label(vcat(*colored_diff_map), "Diff Map"),
        )

        comparison = torch.nn.functional.interpolate(
            comparison.unsqueeze(0), 
            scale_factor=0.5, 
            mode='bicubic', 
            align_corners=False
        ).squeeze(0)
        
        self.logger.log_image(
            "comparison",
            [prep_image(add_border(comparison))],
            step=self.global_step,
            caption=batch["scene"],
        )

        # self.logger.log_image(
        #     key="comparison",
        #     images=[wandb.Image(prep_image(add_border(comparison)), caption=batch["scene"], file_type="jpg")],
        #     step=self.global_step
        # )

        # Render projections and construct projection image.
        # These are disabled for now, since RE10k scenes are effectively unbounded.

        # if isinstance(gaussians, Gaussians):
        #     projections = hcat(
        #             *render_projections(
        #                 gaussians,
        #                 256,
        #                 extra_label="",
        #             )[0]
        #         )
        #     self.logger.log_image(
        #         "projection",
        #         [prep_image(add_border(projections))],
        #         step=self.global_step,
        #     )

        # Draw cameras.
        # cameras = hcat(*render_cameras(batch, 256))
        # self.logger.log_image(
        #     "cameras", [prep_image(add_border(cameras))], step=self.global_step
        # )

        if self.encoder_visualizer is not None:
            for k, image in self.encoder_visualizer.visualize(
                batch["context"], self.global_step
            ).items():
                self.logger.log_image(k, [prep_image(image)], step=self.global_step)
        
        # Run video validation step.
        self.render_video_interpolation(batch)
        self.render_video_wobble(batch)
        if self.train_cfg.extended_visualization:
            self.render_video_interpolation_exaggerated(batch)

    @rank_zero_only
    def render_video_wobble(self, batch: BatchedExample) -> None:
        # Two views are needed to get the wobble radius.
        _, v, _, _ = batch["context"]["extrinsics"].shape
        if v != 2:
            return

        def trajectory_fn(t):
            origin_a = batch["context"]["extrinsics"][:, 0, :3, 3]
            origin_b = batch["context"]["extrinsics"][:, 1, :3, 3]
            delta = (origin_a - origin_b).norm(dim=-1)
            extrinsics = generate_wobble(
                batch["context"]["extrinsics"][:, 0],
                delta * 0.25,
                t,
            )
            intrinsics = repeat(
                batch["context"]["intrinsics"][:, 0],
                "b i j -> b v i j",
                v=t.shape[0],
            )
            return extrinsics, intrinsics

        return self.render_video_generic(batch, trajectory_fn, "wobble", num_frames=60)

    @rank_zero_only
    def render_video_interpolation(self, batch: BatchedExample) -> None:
        _, v, _, _ = batch["context"]["extrinsics"].shape

        def trajectory_fn(t):
            extrinsics = interpolate_extrinsics(
                batch["context"]["extrinsics"][0, 0],
                (
                    batch["context"]["extrinsics"][0, 1]
                    if v == 2
                    else batch["target"]["extrinsics"][0, 0]
                ),
                t,
            )
            intrinsics = interpolate_intrinsics(
                batch["context"]["intrinsics"][0, 0],
                (
                    batch["context"]["intrinsics"][0, 1]
                    if v == 2
                    else batch["target"]["intrinsics"][0, 0]
                ),
                t,
            )
            return extrinsics[None], intrinsics[None]

        return self.render_video_generic(batch, trajectory_fn, "rgb")

    @rank_zero_only
    def render_video_interpolation_exaggerated(self, batch: BatchedExample) -> None:
        # Two views are needed to get the wobble radius.
        _, v, _, _ = batch["context"]["extrinsics"].shape
        if v != 2:
            return

        def trajectory_fn(t):
            origin_a = batch["context"]["extrinsics"][:, 0, :3, 3]
            origin_b = batch["context"]["extrinsics"][:, 1, :3, 3]
            delta = (origin_a - origin_b).norm(dim=-1)
            tf = generate_wobble_transformation(
                delta * 0.5,
                t,
                5,
                scale_radius_with_t=False,
            )
            extrinsics = interpolate_extrinsics(
                batch["context"]["extrinsics"][0, 0],
                (
                    batch["context"]["extrinsics"][0, 1]
                    if v == 2
                    else batch["target"]["extrinsics"][0, 0]
                ),
                t * 5 - 2,
            )
            intrinsics = interpolate_intrinsics(
                batch["context"]["intrinsics"][0, 0],
                (
                    batch["context"]["intrinsics"][0, 1]
                    if v == 2
                    else batch["target"]["intrinsics"][0, 0]
                ),
                t * 5 - 2,
            )
            return extrinsics @ tf, intrinsics[None]

        return self.render_video_generic(
            batch,
            trajectory_fn,
            "interpolation_exagerrated",
            num_frames=300,
            smooth=False,
            loop_reverse=False,
        )

    @rank_zero_only
    def render_video_generic(
        self,
        batch: BatchedExample,
        trajectory_fn: TrajectoryFn,
        name: str,
        num_frames: int = 30,
        smooth: bool = True,
        loop_reverse: bool = True,
    ) -> None:
        # Render probabilistic estimate of scene.
        encoder_output = self.model.encoder((batch["context"]["image"]+1)/2, self.global_step)
        gaussians, pred_pose_enc_list = encoder_output.gaussians, encoder_output.pred_pose_enc_list

        t = torch.linspace(0, 1, num_frames, dtype=torch.float32, device=self.device)
        if smooth:
            t = (torch.cos(torch.pi * (t + 1)) + 1) / 2

        extrinsics, intrinsics = trajectory_fn(t)

        _, _, _, h, w = batch["context"]["image"].shape

        # TODO: Interpolate near and far planes?
        near = repeat(batch["context"]["near"][:, 0], "b -> b v", v=num_frames)
        far = repeat(batch["context"]["far"][:, 0], "b -> b v", v=num_frames)
        output = self.model.decoder.forward(
            gaussians, extrinsics, intrinsics, near, far, (h, w), "depth"
        )
        images = [
            vcat(rgb, depth)
            for rgb, depth in zip(output.color[0], vis_depth_map(output.depth[0]))
        ]

        video = torch.stack(images)
        video = (video.clip(min=0, max=1) * 255).type(torch.uint8).cpu().numpy()
        if loop_reverse:
            video = pack([video, video[::-1][1:-1]], "* c h w")[0]
        visualizations = {
            f"video/{name}": wandb.Video(video[None], fps=30, format="mp4")
        }
            
        # Since the PyTorch Lightning doesn't support video logging, log to wandb directly.
        try:
            wandb.log(visualizations)
        except Exception:
            assert isinstance(self.logger, LocalLogger)
            for key, value in visualizations.items():
                tensor = value._prepare_video(value.data)
                clip = mpy.ImageSequenceClip(list(tensor), fps=30)
                dir = LOG_PATH / key
                dir.mkdir(exist_ok=True, parents=True)
                clip.write_videofile(
                    str(dir / f"{self.global_step:0>6}.mp4"), logger=None
                )

    def print_preview_metrics(self, metrics: dict[str, float | Tensor], methods: list[str] | None = None, overlap_tag: str | None = None) -> None:
        if getattr(self, "running_metrics", None) is None:
            self.running_metrics = metrics
            self.running_metric_steps = 1
        else:
            s = self.running_metric_steps
            self.running_metrics = {
                k: ((s * v) + metrics[k]) / (s + 1)
                for k, v in self.running_metrics.items()
            }
            self.running_metric_steps += 1

        if overlap_tag is not None:
            if getattr(self, "running_metrics_sub", None) is None:
                self.running_metrics_sub = {overlap_tag: metrics}
                self.running_metric_steps_sub = {overlap_tag: 1}
            elif overlap_tag not in self.running_metrics_sub:
                self.running_metrics_sub[overlap_tag] = metrics
                self.running_metric_steps_sub[overlap_tag] = 1
            else:
                s = self.running_metric_steps_sub[overlap_tag]
                self.running_metrics_sub[overlap_tag] = {k: ((s * v) + metrics[k]) / (s + 1)
                                                         for k, v in self.running_metrics_sub[overlap_tag].items()}
                self.running_metric_steps_sub[overlap_tag] += 1

        metric_list = ["psnr", "lpips", "ssim"]

        def print_metrics(runing_metric, methods=None):
            table = []
            if methods is None:
                methods = ['ours']

            for method in methods:
                row = [
                    f"{runing_metric[f'{metric}_{method}']:.3f}"
                    for metric in metric_list
                ]
                table.append((method, *row))

            headers = ["Method"] + metric_list
            table = tabulate(table, headers)
            print(table)

        print("All Pairs:")
        print_metrics(self.running_metrics, methods)
        if overlap_tag is not None:
            for k, v in self.running_metrics_sub.items():
                print(f"Overlap: {k}")
                print_metrics(v, methods)

    def configure_optimizers(self):
        new_params, new_param_names = [], []
        pretrained_params, pretrained_param_names = [], []
        for name, param in self.named_parameters():
            if not param.requires_grad:
                continue
            
            if "gaussian_param_head" in name or "interm" in name:
                new_params.append(param)
                new_param_names.append(name)
            else:
                pretrained_params.append(param)
                pretrained_param_names.append(name)
        
        param_dicts = [
            {
                "params": new_params,
                "lr": self.optimizer_cfg.lr,
             },
            {
                "params": pretrained_params,
                "lr": self.optimizer_cfg.lr * self.optimizer_cfg.backbone_lr_multiplier,
            },
        ]
        optimizer = torch.optim.AdamW(param_dicts, lr=self.optimizer_cfg.lr, weight_decay=0.05, betas=(0.9, 0.95))
        warm_up_steps = self.optimizer_cfg.warm_up_steps
        warm_up = torch.optim.lr_scheduler.LinearLR(
            optimizer,
            1 / warm_up_steps,
            1,
            total_iters=warm_up_steps,
        )
        
        lr_scheduler = torch.optim.lr_scheduler.CosineAnnealingLR(optimizer, T_max=get_cfg()["trainer"]["max_steps"], eta_min=self.optimizer_cfg.lr * 0.1)
        lr_scheduler = torch.optim.lr_scheduler.SequentialLR(optimizer, schedulers=[warm_up, lr_scheduler], milestones=[warm_up_steps])

        return {
            "optimizer": optimizer,
            "lr_scheduler": {
                "scheduler": lr_scheduler,
                "interval": "step",
                "frequency": 1,
            },
        }