mypersonalsharingspot11/artifacts_evaluation

GRADE artifact evaluation (MobiCom 2026)

This anonymous artifact reproduces the quantitative evaluation of the GRADE paper. The release provides 19 model checkpoints plus one auxiliary TAESD weight file (20 .safetensors files in total), code for local inference and metric computation, and the saved camera-ready quantitative results.

Two evaluation paths are supported:

• E1 — saved-result reproduction: regenerate the quantitative tables and figures from the supplied camera-ready CSVs and auxiliary evaluation data. This requires no inference or dataset download.
• E2 — optional local reproduction: run inference on the released evaluation data, compute 2D/3D metrics, then regenerate the quantitative tables and figures from those newly computed CSVs. Choose E2 only when reproducing from the released checkpoints is desired instead of using the supplied pre-computed metric files.

Figure 10 is qualitative-only and is intentionally outside this artifact's quantitative workflow.

Layout

Artifacts_Evaluation/
├── README.md                         # this guide
├── ae.tex                            # artifact-evaluation writeup
├── environment.txt                   # Python 3.11 package specification
├── checkpoints/                      # placeholder for downloaded model weights
├── evaluation_dataset/               # placeholder for downloaded evaluation data
├── src/                              # released model implementations and configs
│   ├── Baselines/                    # DA3, GRT, CaFNet, RadarCam-Depth
│   ├── Ablation/                     # ablation implementations
│   ├── GRADE/                        # Stage 1 radar-depth and Stage 2 diffusion modules
│   └── models/                       # per-model release-relative configurations
└── evaluation/
    ├── run_inference.py              # unified inference runner
    ├── run_metrics.py                # 2D, 3D, merged-CSV, and radar-robustness runner
    ├── reproduce_paper.py            # quantitative paper reproduction
    ├── utils/                         # internal metric-computation implementation
    ├── outputs/inference/            # newly generated predictions
    ├── metric_results/               # newly computed metric CSVs
    ├── reference_results/            # supplied camera-ready CSVs and paper assets
    └── reproduced_results/           # newly generated tables and figures

evaluation/reference_results/ is read-only golden material. The runners write only to evaluation/outputs/, evaluation/metric_results/, and evaluation/reproduced_results/. The package does not include saved inference .npy files.

Model-name mapping

Pass the following identifiers to --model. grade is the preferred name for the complete model; ours_full remains a compatibility alias and historical CSV name.

Code identifier	Paper name
grade / ours_full	GRADE
ours_radar	Ours_radar
ours_diffusion	Ours_diffusion
ours_radar_no_grad	Ours_radar w/o gradient loss
ours_radar_no_doppler	Ours_radar w/o Doppler
ours_full_no_3d	GRADE w/o 3D losses
grt_refine_freeze	GRT refinement (frozen)
grt_refine_retrain	GRT refinement (retrained)
da3	DA3
grt	GRT
grt_image	GRT+Image
grt_no_doppler	GRT w/o Doppler
cafnet	CaFNet
cafnet_no_smoke	CaFNet (No-Smoke)
radarcam-depth	RadarCam-Depth

Step-by-step setup

All three Hugging Face repositories are public and do not require an account or access token. The optional hf command examples require the Hugging Face CLI; if it is not already available before downloading this package, use the linked browser download instead. environment.txt includes huggingface_hub for subsequent CLI use.

1. Prepare the environment. Use Python 3.11 and prepare the required environment from environment.txt. E1 can run on CPU. E2 requires a CUDA-compatible NVIDIA GPU and compatible PyTorch installation.

2. Download the artifact package (about 123 MB). Download artifacts and saved reference results, or use:

   hf download mypersonalsharingspot11/artifacts_evaluation \
     --repo-type dataset --local-dir grade-ae
   cd grade-ae
   

   The downloaded package is sufficient for E1.

3. Run E1, or continue only if E2 is desired. E2 is optional: it is for evaluators who want to create predictions and metrics directly from the released checkpoints rather than reuse the supplied pre-computed metric files.

4. For E2 only, download checkpoints and evaluation data. Download checkpoints and the evaluation dataset, or use the following commands from the artifact root. These downloads are approximately 16 GiB and 176 GiB, respectively.

   hf download mypersonalsharingspot11/checkpoints \
     --repo-type dataset --local-dir checkpoints
   hf download mypersonalsharingspot11/evaluation_dataset \
     --repo-type dataset --local-dir evaluation_dataset
   python -m zipfile -e \
     evaluation_dataset/Smoke-Eval-RadarCam-Depth.zip evaluation_dataset
   

5. Check the E2 layout. The extracted directories must be placed as follows:

   checkpoints/baselines/
   checkpoints/grade/
   checkpoints/ablations/
   checkpoints/misc/
   evaluation_dataset/Smoke-Eval/
   evaluation_dataset/Smoke-Eval-RadarCam-Depth/
   

   All models except radarcam-depth use Smoke-Eval. If extraction creates an extra directory level, move the inner Smoke-Eval-RadarCam-Depth/ directory up so that it sits directly under evaluation_dataset/.

E1: reproduce the saved camera-ready results

Run:

python evaluation/reproduce_paper.py --mode saved

The output is written to evaluation/reproduced_results/saved/. It contains Tables 2–7 and the panels for Figures 9 and 11–14. As quick checks, Table 2 reports GRADE clear/smoke MAE values of 0.303/0.313, Table 3 reports 0.295/0.304 for light/heavy smoke, Table 4 reports 0.308/0.320/0.360 for full/radar/diffusion, and Table 7 identifies eight DDIM steps as best.

E2 (optional): reproduce results from local inference

This path is optional and is not needed to reproduce the supplied camera-ready results. Use it to independently regenerate predictions, metric CSVs, tables, and figures from the checkpoints and evaluation dataset.

For each desired model, run the inference, metric, and paper-reproduction runners in that order:

python evaluation/run_inference.py --model grade --gpuid 0
python evaluation/run_metrics.py --model grade --workers 8
python evaluation/reproduce_paper.py --mode local

The inference runner launches through Accelerate by default and accepts direct accelerate launch invocation as well. Repeat the first two commands for every model needed by the desired table or figure before the final command. GRADE diffusion variants use the released fixed DDIM seed (42); each frame receives a reproducible, newly drawn noise sample from that seeded generator.

To reproduce the radar-robustness analysis after generating the prerequisite predictions and metric outputs, run:

python evaluation/run_metrics.py --radar-robustness

--mode local writes to evaluation/reproduced_results/local/. A complete GPU-based E2 run over all models is substantially more expensive than E1; provision roughly 400 GiB of free storage and expect the full workflow to take longer than a day on typical local hardware.

Useful options and checks

• All three runners provide --dry-run to validate argument dispatch without running a model.
• run_inference.py refuses to overwrite existing predictions. Use a different output root when retaining an earlier run.
• run_metrics.py supports --sequence, --target-resolution, and --workers for smaller functional checks; its default worker count is 1 and it caps workers at 8.
• reproduce_paper.py supports --tables-only and --figures-only.

Changing the data subset, output resolution, sampling settings, or evaluation configuration is useful for a smoke test but will not reproduce the camera-ready numerical values exactly.