You can deploy this model and start earning money today!

Chahine-Nicolas/
MoE-DSI-3D

Name: MoE-DSI-3D
Availability: InStock

Views

No views yet

Models for the paper "Improving retrieval in large-scale 3D point clouds" (submitted for publication)

This repository provides the weights for MoE-DSI-3D, including the expert models (git_hilbert_xy.zip), the gating model (Gate_Model.zip), a monolithic model trained on Paris_East (git_hilbert_EAST.zip), and the updated tokens vocabulary for git-base-coco (transformers_vocab.zip).

MoE-DSI-3D Checkpoint

Download here

Paris_East Experts

Expert	File
A	`git_hilbert_A0.zip`
B	`git_hilbert_B0.zip`
C	`git_hilbert_C0.zip`
D	`git_hilbert_D0.zip`

Paris_West Experts

Expert	File
A	`git_hilbert_A1.zip`
B	`git_hilbert_B1.zip`
C	`git_hilbert_C1.zip`
D	`git_hilbert_D1.zip`
E	`git_hilbert_E1.zip`

`Gate_Model.zip`:

Model	Description
`gate_east`	Routes inputs to the Paris_East experts.
`gate_west`	Routes inputs to the Paris_West experts.
`gate_east_west`	Selects between the Paris_East and Paris_West expert groups.

Fine-tuned LoGG3D-Net on LiDAR HD.

LoGG3D-Net (Re-trained on LHD).zip

Tokenizer

Tokenizer vocabulary optimized for Hilbert-curve indexing of LiDAR HD.

transformers_vocab.zip

How to use

Refer to the source code.

Chahine-Nicolas/MoE-DSI-3D