Skip to content
Other

RCQFusion

3D object detection from camera and radar data

2 min readupdated

A model for 3D object detection in driving scenes that fuses camera and radar data (Camera–Radar 3D Detection). It outputs 3D coordinates and class predictions across three categories: Pedestrian, Cyclist, and Car. Built on MMDetection3D (a library for 3D object detection), the solution was released as part of an anonymous ICLR submission. Unlike vision-only detectors, it uses radar data and temporal links across frames to more accurately estimate the distance and position of road users. Useful for autonomous driving researchers and developers looking to reproduce experiments on the View of Delft (VoD) dataset.

Key features

  • Camera and 5-frame radar modality fusion accounting for temporal relationships across sequence frames.
  • Detection of three core road object classes: Pedestrian, Cyclist, and Car.
  • Complete inference and validation pipeline included: a ready-to-use checkpoint, VoD data conversion scripts, and a bundled version of MMDetection3D 1.0.0rc6.
  • Single- and multi-GPU execution support with half-precision (FP16) computation.

What you can do

  • Evaluate 3D detection performance on the View of Delft validation split using the official AP11 metric.
  • Use the provided checkpoint and data preparation scripts as a baseline for benchmarking your own camera–radar fusion algorithms.
  • Run distributed inference across multiple GPUs to speed up generating 3D predictions for driving scenes.

Example

export VOD_DATA_ROOT=/path/to/vod
CUDA_VISIBLE_DEVICES=0 bash tools/infer_vod.sh
Read moreDetails
Validation (1,296 VoD frames, 3D AP11): overall mAP — 61.35% (Pedestrian 53.64%, Cyclist 76.23%, Car 54.18%), driving corridor (ROI) mAP — 77.92% (Cyclist up to 90.84%).
Environment: Linux, Python 3.8, PyTorch 1.9.1+cu111, mmcv-full 1.6.0 with CUDA operators; tested on an NVIDIA RTX A5000.
Limitations: the VoD dataset itself is not included in the repository and must be downloaded separately under a research license.
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 0.2 GB · ready files0.7 GB
  • FP8 / INT8recommendedweights 0.1 GB0.6 GB
  • 4-bitweights 0.1 GB0.6 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · BF16 — from RTX 2060

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # iclr-submission-58778/rcqfusion-vod
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download iclr-submission-58778/rcqfusion-vod --local-dir ~/models/rcqfusion-vod
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to rcqfusion-vod
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of rcqfusion-vod (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

iclr-submission-58778/rcqfusion-vod
Author
iclr-submission-58778
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments

Other

mars-dfd-hr

mars-dfd-hr