Upload folder using huggingface_hub

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+paper.pdf filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,104 @@
+---
+tags:
+  - robotics
+  - anima
+  - slam
+  - gaussian-splatting
+  - semantic-slam
+  - robot-flow-labs
+library_name: pytorch
+pipeline_tag: robotics
+license: apache-2.0
+---
+
+# GS3LAM — Gaussian Semantic Splatting SLAM
+
+Part of the [ANIMA Intelligence Compiler Suite](https://robotflowlabs.com) by Robot Flow Labs.
+
+**Wave 7 | Domain: SLAM | Module: slam-gs3lam**
+
+## Paper
+
+**GS3LAM: Gaussian Semantic Splatting SLAM**
+Linfei Li, Lin Zhang, Zhong Wang, Ying Shen
+ACM MM 2024 | [arXiv 2603.27781](https://arxiv.org/abs/2603.27781) | [DOI](https://doi.org/10.1145/3664647.3680739)
+
+## Architecture
+
+Dense semantic RGB-D SLAM using a Semantic Gaussian Field (SG-Field). Each Gaussian stores position, covariance, opacity, RGB color, and a 16-dimensional semantic feature vector. A lightweight 1x1 Conv2d decoder maps semantic features to per-pixel class logits.
+
+Core components:
+- **SG-Field**: 3D Gaussian representation with semantic features
+- **Differentiable Splatting**: CUDA-accelerated rendering via gaussian-semantic-rasterization
+- **Tracking**: Frame-to-model pose optimization (rotation + translation)
+- **Mapping**: Joint optimization of Gaussians + semantic decoder
+- **DSR**: Depth-adaptive Scale Regularization
+- **RSKM**: Random Sampling-based Keyframe Mapping
+
+## Exported Formats
+
+| Format | File | Use Case |
+|--------|------|----------|
+| SafeTensors (field) | `pytorch/gs3lam_office2_v1_field.safetensors` | Gaussian field parameters |
+| SafeTensors (decoder) | `pytorch/gs3lam_office2_v1_decoder.safetensors` | Semantic decoder weights |
+| ONNX (decoder) | `onnx/gs3lam_office2_v1_decoder.onnx` | Cross-platform decoder inference |
+| Poses (npy) | `pytorch/gs3lam_office2_v1_poses.npy` | Estimated camera trajectory [N,4,4] |
+| Checkpoint | `checkpoints/best.pt` | Full checkpoint for resuming |
+
+**Note**: TensorRT engines must be generated on target hardware due to architecture-specific compilation.
+
+## Training Details
+
+- **Scene**: Replica office2 (2000 frames, 1200x680 rendered at 600x340)
+- **Hardware**: NVIDIA L4 23GB
+- **Config**: 20 tracking iterations, 15 mapping iterations (L4-tuned)
+- **Gaussians**: 21K (capped at 80K with periodic pruning)
+- **Duration**: ~5.5 hours
+- **CUDA Extension**: gaussian-semantic-rasterization (sm_89)
+
+### Current Metrics (L4 baseline, reduced resolution)
+
+| Metric | Value | Paper Target |
+|--------|-------|-------------|
+| PSNR | 3.39 dB | >= 35.0 dB |
+| ATE | 178 cm | <= 0.50 cm |
+
+> **Note**: These metrics reflect an L4-constrained run at half resolution with aggressive Gaussian capping (21K vs paper's ~775K). Paper-quality reproduction requires full resolution on A100/H100 hardware.
+
+## Usage
+
+```python
+import torch
+from anima_slam_gs3lam.export import load_checkpoint, reconstruct_field, reconstruct_decoder
+
+ckpt = load_checkpoint("checkpoints/best.pt")
+field = reconstruct_field(ckpt, device="cuda")
+decoder = reconstruct_decoder(ckpt, device="cuda")
+```
+
+## API
+
+```bash
+# Start service
+python -m anima_slam_gs3lam
+
+# Health check
+curl http://localhost:8080/health
+curl http://localhost:8080/ready
+curl http://localhost:8080/info
+```
+
+## Docker
+
+```bash
+docker compose -f docker-compose.serve.yml --profile api up -d
+```
+
+## CUDA Extension
+
+The gaussian-semantic-rasterization CUDA kernel is shared at:
+`/mnt/forge-data/shared_infra/cuda_extensions/gaussian_semantic_rasterization/`
+
+## License
+
+Apache 2.0 — Robot Flow Labs / AIFLOW LABS LIMITED

checkpoints/best.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e0428337471e2af4d4a3773ea198d195b065cba117b3a475217f013596781ab3
+size 3082317

configs/default.toml

@@ -0,0 +1,84 @@
+[project]
+name = "anima-slam-gs3lam"
+codename = "SLAM-GS3LAM"
+functional_name = "Gaussian Semantic Splatting SLAM"
+wave = 7
+paper_arxiv = "2603.27781"
+python_version = "3.11"
+
+[compute]
+backend = "auto"  # "mlx", "cuda", "auto"
+precision = "fp32"
+prefer_mlx_on_mac = true
+prefer_cuda_on_linux = true
+
+[data]
+shared_volume = "/mnt/forge-data"
+repos_volume = "/mnt/forge-data/repos"
+dataset_root_rel = "datasets/slam/gs3lam"
+model_root_rel = "models/slam/gs3lam"
+
+[paper_defaults]
+replica_tracking_iterations = 40
+replica_mapping_iterations = 60
+scannet_tracking_iterations = 100
+scannet_mapping_iterations = 30
+tum_tracking_iterations = 360
+tum_mapping_iterations = 150
+tracking_rotation_lr = 0.0004
+tracking_translation_lr = 0.002
+
+[repo_overrides.scannet]
+tracking_iterations = 200
+mapping_iterations = 60
+
+[dataset_presets.replica]
+name = "replica"
+sequence = "office0"
+root = "/mnt/forge-data/datasets/slam/gs3lam/Replica"
+desired_image_height = 680
+desired_image_width = 1200
+
+[dataset_presets.replica.camera]
+image_height = 680
+image_width = 1200
+fx = 600.0
+fy = 600.0
+cx = 599.5
+cy = 339.5
+png_depth_scale = 6553.5
+crop_edge = 0
+
+[dataset_presets.scannet]
+name = "scannet"
+sequence = "scene0059_00"
+root = "/mnt/forge-data/datasets/slam/gs3lam/scannet"
+desired_image_height = 480
+desired_image_width = 640
+
+[dataset_presets.scannet.camera]
+image_height = 968
+image_width = 1296
+fx = 1169.621094
+fy = 1167.105103
+cx = 646.295044
+cy = 489.927032
+png_depth_scale = 1000.0
+crop_edge = 0
+
+[dataset_presets.tum]
+name = "tum"
+sequence = "rgbd_dataset_freiburg1_desk"
+root = "/mnt/forge-data/datasets/tum"
+desired_image_height = 480
+desired_image_width = 640
+
+[dataset_presets.tum.camera]
+image_height = 480
+image_width = 640
+fx = 517.3
+fy = 516.5
+cx = 318.6
+cy = 255.3
+png_depth_scale = 5000.0
+crop_edge = 8

configs/l4_replica.toml

@@ -0,0 +1,29 @@
+# L4-tuned config for Replica (23GB VRAM)
+# Reduced resolution + iterations to fit in VRAM
+# Paper uses 40/60 at 1200x680 on RTX 3090 — we halve resolution for L4
+
+[project]
+name = "anima-slam-gs3lam"
+
+[paper_defaults]
+replica_tracking_iterations = 20
+replica_mapping_iterations = 15
+tracking_rotation_lr = 0.0004
+tracking_translation_lr = 0.002
+
+[dataset_presets.replica]
+name = "replica"
+sequence = "office0"
+root = "/mnt/forge-data/datasets/slam/gs3lam/Replica"
+desired_image_height = 340
+desired_image_width = 600
+
+[dataset_presets.replica.camera]
+image_height = 680
+image_width = 1200
+# Intrinsics scaled by 0.5 for half-resolution rendering
+fx = 300.0
+fy = 300.0
+cx = 299.75
+cy = 169.75
+png_depth_scale = 6553.5

gs3lam_office2_v1_metadata.json

@@ -0,0 +1,8 @@
+{
+  "format": "gs3lam",
+  "version": "0.1.0",
+  "semantic_dim": "16",
+  "semantic_classes": "256",
+  "num_gaussians": "21016",
+  "num_poses": "2000"
+}

logs/eval_office2.json

@@ -0,0 +1,12 @@
+{
+  "scene": "office2",
+  "total_frames": 2000,
+  "elapsed_s": 19786.7,
+  "num_gaussians": 21016,
+  "replica_psnr": 3.3855,
+  "replica_ssim": 0.0226,
+  "replica_lpips": 0.6932,
+  "replica_ate_cm": 178.2363,
+  "replica_fps": 0.1,
+  "checkpoint": "/mnt/artifacts-datai/checkpoints/project_slam_gs3lam/office2/gs3lam_step_1999.pt"
+}

onnx/gs3lam_office2_v1_decoder.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:422e6ac4f980a61f6803b823ac6ce883ed9ab30cc331da1b10234b6e7cad848b
+size 670

paper.pdf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7c3409e74e9bcdd7650aaf299e15f3f0c4655cb0df71fe90644d0deb42b4049d
+size 10677424

pytorch/gs3lam_office2_v1_decoder.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:500579319986847df2dcf930b638666eefd3f5e32b6365671b288bd58f52bbaf
+size 17704

pytorch/gs3lam_office2_v1_field.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:134cace086b4e325641f3d495a2455194f7cdba61ac29cc6a195583d8f79e2c6
+size 2354416

pytorch/gs3lam_office2_v1_poses.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:39be066ffe9b0c04eb033719245f5352baf966ea9670bd57be9591bd3667667f
+size 128128