license: apache-2.0
base_model: facebook/EUPE-ViT-B
tags:
- depth-estimation
- frozen-backbone
- vision-transformer
- architecture-search
library_name: pytorch
datasets:
- sayakpaul/nyu_depth_v2
Depth Heads
A systematic study of monocular depth head architectures operating on frozen vision transformer features. Given a dense spatial feature grid from any frozen backbone, what is the most parameter-efficient architecture for per-pixel metric depth prediction?
Standard practice treats the backbone and depth decoder as a joint system. Recent universal encoders produce spatial features of sufficient quality that the backbone can remain frozen while a lightweight head is trained on depth data. Under this regime, the head is the only variable.
This repository contains an arena framework for rapid comparison of depth head candidates and a collection of architectures spanning conventional decoders through novel minimal-parameter designs. All heads consume the same spatial feature tensor and produce per-pixel depth maps. The reference backbone is EUPE-ViT-B (86M parameters, frozen), but the framework is backbone-agnostic — the same heads can be evaluated against any frozen ViT that produces a stride-16 spatial feature grid.
Heads
Nine architectures, all consuming a [B, 768, H, W] spatial feature tensor and producing [B, 1, H_out, W_out] metric depth in meters over the 0.001–10 range. Each head lives in its own folder under heads/ with a single head.py implementation.
| Name | Architecture | Parameters |
|---|---|---|
linear_probe |
BatchNorm + 1×1 conv → 256 depth bins, weighted-sum decode. The EUPE paper baseline. | ~199K |
cofiber_linear |
Cofiber decomposition + shared 1×1 conv per scale → 256-bin decode | ~197K |
cofiber_threshold |
Cofiber decomposition + per-scale LayerNorm + prototype prediction → 256-bin decode | ~202K |
wavelet |
Haar wavelet decomposition + per-subband prediction → 256-bin decode | ~590K |
log_linear |
Single 1×1 conv predicting log-depth, exponentiated and clamped | 769 |
ordinal_regression |
K independent threshold classifiers, depth = sum of positive predictions × bin width | ~49K |
multiscale_gradient |
Per-scale depth gradient prediction on cofiber bands, integrated for absolute depth | ~6K |
harmonic |
Cofiber edge detection + boundary depth prediction + Jacobi Laplace solve at non-edge locations | 770 |
renormalization |
Depth from per-scale cofiber energy weighted sum (one weight per scale) | 6 |
Arena Framework
arena.py runs any head by name against cached NYU Depth V2 backbone features. The arena pre-extracts features once, then each candidate trains and evaluates without touching the backbone again. Training is SILog loss at 416×416 resolution against the indoor depth label space; evaluation reports root mean squared error (RMSE) on the test split.
Status
Heads are implemented and importable through the heads/ registry. The arena screening sweep across all 9 heads has not yet been run on a fresh NYU Depth V2 cache; results will be published here when available.