Idol weights (FP32) and model card

README.md

@@ -0,0 +1,45 @@
+---
+license: apache-2.0
+library_name: idol
+pipeline_tag: image-to-3d
+tags:
+- gaussian-splatting
+- 3d
+- image-to-3d
+---
+
+# Idol
+
+**A single image to a 3D object.** Idol turns one picture of an object into a 3D Gaussian splat (up to
+1,048,576 Gaussians). Code and inference: [github.com/speridlabs/idol-sdk](https://github.com/speridlabs/idol-sdk) ·
+Demo: [Hugging Face Space](https://huggingface.co/spaces/speridlabs/idol).
+
+## Files
+
+| file | |
+|---|---|
+| `denoiser.safetensors` | rectified-flow transformer, 370M parameters, EMA weights, FP32 |
+| `decoder.safetensors` | octree decoder that places the anchor points, 51M parameters, FP32 |
+| `decoder_gs.safetensors` | Gaussian decoder, 32 Gaussians per anchor, 237M parameters, FP32 |
+| `birefnet.safetensors` | BiRefNet background remover |
+| `config.json` | model configs and inference defaults |
+
+## Usage
+
+```bash
+pip install git+https://github.com/speridlabs/idol-sdk
+python -m idol.sample --image photo.jpg --mode real
+```
+
+```python
+from idol import IdolPipeline
+
+pipeline = IdolPipeline.from_pretrained("speridlabs/idol")
+output = pipeline("photo.jpg", mode="real")     # mode="synthetic" for renders and generated images
+output.gaussian.save_ply("photo.ply")
+```
+
+The image encoders, [DINOv3](https://huggingface.co/facebook/dinov3-vith16plus-pretrain-lvd1689m) and the
+[FLUX.2](https://huggingface.co/black-forest-labs/FLUX.2-dev) VAE, are gated: request access, then `hf auth login`.
+
+Defaults: seed 42, 50 Euler steps, CFG 3.0, time shift 3.0, 262,144 Gaussians.

birefnet.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9ab37426bf4de0567af6b5d21b16151357149139362e6e8992021b8ce356a154
+size 444473596

config.json

@@ -0,0 +1,89 @@
+{
+  "denoiser": {
+    "name": "LatentSeqMMFlowModel",
+    "args": {
+      "q_token_length": 8192,
+      "in_channels": 16,
+      "cam_channels": 5,
+      "out_channels": 16,
+      "model_channels": 1024,
+      "cond_channels": 1280,
+      "cond2_channels": 128,
+      "num_refiner_blocks": 2,
+      "num_blocks": 24,
+      "num_heads": 16,
+      "mlp_ratio": 4,
+      "pe_mode": "3d_ot_repo",
+      "qk_rms_norm": true,
+      "share_mod": true,
+      "use_shift_table": true,
+      "use_fp16": true
+    }
+  },
+  "decoder": {
+    "name": "OctreeProbabilityFixedlenDecoder",
+    "args": {
+      "pcd_pe_mode": "pcd_ape_v2",
+      "model_channels": 1024,
+      "cond_channels": 16,
+      "num_blocks": 4,
+      "num_heads": 16,
+      "mlp_ratio": 4,
+      "share_mod": true,
+      "use_fp16": true
+    }
+  },
+  "decoder_gs": {
+    "name": "GaussianFixedlenDecoder",
+    "args": {
+      "pcd_pe_mode": "pcd_ape_v2",
+      "in_channels": 3,
+      "model_channels": 1024,
+      "cond_channels": 16,
+      "num_blocks": 16,
+      "num_heads": 16,
+      "mlp_ratio": 4,
+      "use_fp16": true,
+      "use_learned_offset_scale": true,
+      "use_per_offset": true,
+      "representation_config": {
+        "lr": {
+          "_xyz": 1.0,
+          "_features_dc": 1.0,
+          "_opacity": 1.0,
+          "_scaling": 1.0,
+          "_rotation": 0.1
+        },
+        "perturb_offset": true,
+        "perturbe_size": 1.5,
+        "offset_scale": 0.05,
+        "num_gaussians": 32,
+        "2d_filter_kernel_size": 0.1,
+        "3d_filter_kernel_size": 0.0009,
+        "scaling_bias": 0.004,
+        "opacity_bias": 0.1,
+        "scaling_activation": "softplus"
+      }
+    }
+  },
+  "image_size": 1024,
+  "max_voxel_level": 8,
+  "max_sampled_points": 8192,
+  "sigma_min": 1e-05,
+  "image_encoders": {
+    "dinov3": "facebook/dinov3-vith16plus-pretrain-lvd1689m",
+    "flux2_vae": "black-forest-labs/FLUX.2-dev",
+    "resolution": 1024
+  },
+  "defaults": {
+    "seed": 42,
+    "steps": 50,
+    "cfg_strength": 3.0,
+    "rescale_t": 3.0,
+    "num_gaussians": 262144,
+    "crop_margin": {
+      "synthetic": 1.2,
+      "real": 1.7
+    }
+  }
+}

decoder.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9da4d7676bfcadb1c3fdfbff796e6257dadf82d26ec05be7717c79b2b13a1c4e
+size 203171696

decoder_gs.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3087eb8ee5763206a53e4078acf2a1bf382dab5147d407ec13a2cf39c04cc544
+size 949083036

denoiser.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ded2cacf671bf3978ebc62729087abdbd318ef4752e6805b4c9fc8ffa51fe716
+size 1482258308