For the first update in a little more than two months, a flurry of updates was pushed into the Nerfstudio codebase. This update mainly features quality of life improvements to the codebase.

One of my favorite additions to 1.1.5 is to add more naming information when exporting plys. I have found it a little difficult to keep them organized on export, making this a welcome addition. Interestingly, there is now the ability to export out a point cloud from their implementation of Zip-NeRF, though I'm not exactly sure how that would work or what that implies.

The team has also bumped versions for dependencies like viser (to 0.2.7) and gsplat (to 1.4.0), rounding out a suite of improvements that culminate in the codebase now reaching version 1.1.5.

Please see below for the full changelog:

• Suppress wandb type error by @brentyi in #3397

• Bump viser to 0.2.7 by @ginazhouhuiwu in #3398

• change grid resolution type by @hoanhle in #3400

• Add eval data to image_rename_map by @jiaruiyu99 in #3396

• update documentation for instant-ngp by @hoanhle in #3406

• Update dockerfile and add docker build action by @jkulhanek in #3283

• RGBA renderer for splatfacto by @hardikdava in #3307

• bugfix: handle multiple camera types in a single batch by @decrispell in #3415

• Add option to start training paused by @aayushg55 in #3420

• Fix docker attestation fail on forked repo PRs by @jkulhanek in #3423

• use gsplat strategy interface to simplify splatfacto by @jb-ye in #3376

• Minor update to generate splat files in inference mode by @maxwildersmith in #3432

• Fix crop obb masking in pointcloud generation by @nepfaff in #3433

• Adding Random Seed for Frame Processing by @Anthony-Tafoya in #3416

• Fix splatfacto crash in eval when using viewer by @aayushg55 in #3430

• Zipnerf point cloud and tsdf mesh export support by @aayushg55 in #3429

• Fix #3417 by @jb-ye in #3441

• Change depth from float64 to float32 by @nepfaff in #3453

• Speed up PairPixelSampler by @nepfaff in #3452

• Rejection sample mask for patch pixel sampler by @nepfaff in #3457

• Make PixelSampler the depth-nerfacto default by @nepfaff in #3454

• Keep depth_image on CPU by @nepfaff in #3449

• Add max_frames + max_output_size arguments for Aria data processing by @brentyi in #3466

• Add option for refining the TSDF using the mesh's AABB by @nepfaff in #3459

• Add option to change .ply output filename for Gaussian exports by @brentyi in #3467

• tag exported ply files by @simonbethke in #3472

• update gsplat version to >=1.3.0 by @kerrj in #3478

• Properly reset train state after render lock in viewer by @kerrj in #3486

• Restrict pymeshlab version on Windows (3461) by @SharkWipf in #3474

• fix: saving ckpt error when ckpt folder has some folder inside by @Crescent-Saturn in #3516

• bump version to 1.1.5 by @jb-ye in #3521

Nerfstudio remains completely free to use and commercially viable with an Apache 2.0 license. If you are considering making a PR, they are welcomed!