← Research Timeline Aditya Jain / Apple Maps · 3D Reconstruction
Apr 2026
Topic 45 Apr 2026 Hypernetworks · SIREN · Weight-Space Latent Codes

Hypernet → Shape —
The Weight-Space Pipeline.

The standalone hypernet-to-shape pipeline — the predecessor that the twelve-phase Hypernet → DeepSDF archive (Topic 41) grew out of and ultimately abandoned. The chain: 24 images of an object → 24 image-SIRENs → one hypernetwork per object; then the hypernet weights → a tiny mapper → a 128-dim latent → an autoencoder decoder → shape-SIREN weights → SDF → mesh. The load-bearing design choice documented here: a weight-space autoencoder rescues mesh quality versus predicting all 264 K shape-SIREN weights directly — the move that the larger Topic-41 archive later diagnoses, and pivots away from, in detail.

00 — Motivation

Shapes as weights — the hypothesis, in its first concrete form.

This is where the "a 3-D shape is the weights of a neural network" hypothesis first becomes a running pipeline. Each object is rendered from 24 viewpoints; each view is overfitted into its own image-SIREN; a hypernetwork is trained per object to map across those 24 image-networks; and the hypernet's weights are then routed through a mapper and an autoencoder into the weights of a shape-SIREN whose SDF, marching-cubed, is the reconstructed mesh. The whole thing is a bet that weight-space relationships between image-networks and geometry-networks can be learned.

The honest framing: this pipeline is the prior work — referenced in the Topic-41 thesis as "our prior work / Topic 03". It established two things that the bigger archive then built on. One, that per-layer hypernetworks can preserve genus-1 topology when reconstructing per-shape MLP weights, while monolithic hypernets destroy topology even at MSE ~10⁻⁷. Two, that warm-starting all shape-SIRENs from a single anchor is necessary for coherent weight-space interpolation. Both findings turn out to be double-edged, as Topic 41 documents.

What it informs
Topic 45 is the direct predecessor of the Hypernet → DeepSDF archive (Topic 41). The weight-space autoencoder here is the "phase 10" of that archive; the warm-start requirement here is the cause of the "warm-start dominance problem" diagnosed there. Reading this page first makes the twelve-phase archive read as what it is — the systematic post-mortem of the pipeline on this page, plus the DeepSDF pivot that finally works.
Pipeline

24 images → 24 image-SIRENs → hypernet → mapper → latent → AE → shape-SIREN.

24 imagesper object 24 image-SIRENsone per view hypernet~17.9M, per object tiny mapper→ latent 128-dim latentAE bottleneck AE decoder→ shape-SIREN wts SDF → mesh264K wts → marching cubes The weight-space autoencoder (orange) rescues mesh quality vs. predicting all 264K shape-SIREN weights directly with a mapper alone. Checkpoints — anchors, autoencoder, mappers, watertight meshes, SDF samples, shape-SIRENs, image-SIRENs, hypernets — on HuggingFace.
01 — The Pipeline Stages

A numbered pipeline, 01_watertight through 80_train_shape_sirens.

The repository is a numbered pipeline — 01_watertight.py through 80_train_shape_sirens.py — plus experiment scripts that branch off it. The stages, in order: download Objaverse objects; watertight-convert; render 24 views per object; train an image-SIREN per view; train a per-object hypernetwork across the 24 image-SIRENs; train the shape-SIRENs. The two competing routes from hypernet weights to shape-SIREN weights are where the design decisions live.

RouteScriptWhat it does
Direct mapper (baseline)hypernet_to_shape_mapper.pyA mapper predicts all 264 K shape-SIREN weights directly from the hypernet weights. Topologically fragile — small weight errors destroy mesh structure
Latent autoencoder (current)autoencoder_pipeline_n100_mlp.pyA weight-space autoencoder compresses to a 128-dim latent first; a tiny mapper targets the latent; the AE decoder reconstructs the weights. Rescues mesh quality
Scaling orchestratorscale_to_n100.pyOrchestrates the scaling experiment to N = 100 objects
OOD testood_test_full.pyOut-of-distribution generalisation test on a held-out shape

Configuration lives in configs/ (CFG.data, CFG.shape_siren, …); core modules — siren, hypernet, render, watertight — in src/. Set CFG.data.num_objects = 100 and run 00_download_objaverse.py80_train_shape_sirens.py in order to reproduce the N = 100 run.

Core Design Choice

Don't predict 264K weights directly. Compress to 128 dims first.
The weight-space autoencoder is the rescue — and, later, the trap.

Predicting all 264 K shape-SIREN weights directly with a mapper is topologically fragile: a small aggregate weight error lands adversarially on the dimensions that control the SDF zero-crossing, and the mesh breaks. The weight-space autoencoder compresses the weights to a 128-dim latent the mapper can actually hit. It works here — and Topic 41 then shows, at larger scale, exactly where and why the autoencoder rescue itself fails: numerical reconstruction cosine is not mesh quality.

02 — What It Established

Two findings that the Topic-41 archive both builds on and unwinds.

Per-layer hypernetworks preserve topology; monolithic ones do not. A per-layer hypernetwork — one that generates the weights of each layer of the shape-SIREN with a dedicated head — preserves genus-1 topology in the reconstructed mesh. A monolithic hypernetwork that emits the whole weight vector at once destroys topology even when the weight-space MSE is driven down to ~10⁻⁷. This is the first appearance of the thesis-line refrain that aggregate weight MSE is a poor proxy for mesh quality — a lesson Topic 41 then states as a hard rule.

Warm-starting from a shared anchor is necessary for coherent weight-space interpolation. If every shape-SIREN is trained independently, the resulting weight vectors live in arbitrary permutation neighbourhoods, and the line segment between two of them in weight space decodes to garbage. Warm-starting all shape-SIRENs from a single anchor keeps them in the same neighbourhood, so interpolation stays on the shape manifold. This pipeline depends on that property — and Topic 41's "warm-start dominance problem" is the discovery that the same property, at the scale of image-conditioned diffusion, is fatal: it concentrates the weight distribution into a thin shell where per-shape signal is unrecoverable.

The honest relationship to Topic 41
This pipeline is not superseded so much as diagnosed. The Hypernet → DeepSDF archive is the systematic study of why this weight-space approach does not extend to image-conditioned generation at the 976-shape scale — and the pivot, to a DeepSDF shared decoder with a constructed 64-dim latent, that finally works. Topic 45 is the thing being post-mortemed; it is on the timeline because the post-mortem only makes sense if you can see the original.

Interactive Demo · Live

Compare the two routes from hypernet to shape. Toggle between the direct-mapper baseline (predict 264 K weights) and the autoencoder route (compress to 128 dims first). The left pane shows the weight-error landing pattern; the right pane shows the resulting mesh quality. Direct prediction lands error on the topology-critical dimensions; the AE route does not.

01 — Route DIRECT MAPPER
02 — Weight-error landing — MSE
03 — Reconstructed mesh

Full Technical Paper

White paper · the hypernet-to-shape weight-space pipeline · per-layer vs monolithic hypernets · the autoencoder rescue · the warm-start requirement · how it sets up the Topic-41 post-mortem

Read Paper →
Related Thesis Chapters
Hypernet → DeepSDF
The successor — the twelve-phase archive that systematically diagnoses why this weight-space pipeline does not extend to image-conditioned generation, and pivots to DeepSDF.
Activation-Space SDF
The segmentation counterpart — also probes what a trained per-shape SDF network has actually encoded in its weights / activations.
SDF Research
The foundational SDF study — the implicit-surface representation the shape-SIRENs in this pipeline reconstruct into.
Appendix — Raw Materials
Transcripts & Source References
████████████████████████████████████████████████

██████████████████████████████████████
█████████ · ████ · █████████████████████
████████████████████████████████████████████████████████████████████████████████████████████████████████████
Restricted Access