Skip to content
Python

update readme

mohamdev8bb8cab8db12
diff --git a/.gitignore b/.gitignore
index 39052df..72fb3f4 100644
--- a/.gitignore
+++ b/.gitignore
@@ -21,6 +21,7 @@ rf-detr*.pth
*.pth.lock
*.onnx
*.pth
+*.mp4
# Rerun recordings
*.rrd
diff --git a/README.md b/README.md
index 5b8dafb..9d3125c 100644
--- a/README.md
+++ b/README.md
@@ -1,5 +1,7 @@
# InstantHMR
+![InstantHMR demo](models/instanthmr.gif)
+
A lightweight, ONNX-exportable distillation of
[`facebook/sam-3d-body-dinov3`](https://huggingface.co/facebook/sam-3d-body-dinov3)
for **3D human pose estimation and mesh recovery**: a RepViT-M1.5 backbone + a 9-token
diff --git a/docs/architecture.md b/docs/architecture.md
index 4017009..21613f9 100644
--- a/docs/architecture.md
+++ b/docs/architecture.md
@@ -75,7 +75,7 @@ The DINOv3-based teacher is excellent but heavy: its tokenizer + decoder runs
at <20 FPS on a desktop GPU and doesn't fit comfortably on edge hardware.
Distillation gives us InstantHMR — a model that:
-- runs at ~150 FPS on a single RTX 4070,
+- runs at ~200 FPS on a single RTX 4070,
- exports to ONNX / TFLite / QNN, and
- preserves the SAM3D 70-keypoint & MHR outputs, so any downstream code that
consumed `pred_keypoints_3d` keeps working.
diff --git a/models/README.md b/models/README.md
index 1434ee4..a947d37 100644
--- a/models/README.md
+++ b/models/README.md
@@ -1,5 +1,25 @@
+---
+license: other
+license_name: sam-license
+license_link: https://github.com/facebookresearch/sam-3d-body/blob/main/LICENSE
+library_name: onnx
+pipeline_tag: keypoint-detection
+tags:
+ - human-pose-estimation
+ - human-mesh-recovery
+ - 3d-pose-estimation
+ - onnx
+ - distillation
+ - sam3d
+ - repvit
+ - cliff
+base_model: facebook/sam-3d-body-dinov3
+---
+
# InstantHMR — model weights
+![InstantHMR demo](instanthmr.gif)
+
A lightweight, ONNX-exportable distillation of
[`facebook/sam-3d-body-dinov3`](https://huggingface.co/facebook/sam-3d-body-dinov3)
for **3D human pose estimation and mesh recovery**: a RepViT-M1.5
@@ -8,7 +28,7 @@ Trained to mimic the SAM3D teacher's per-person 70-keypoint outputs from
a single 224×224 crop.
Demo, training, and inference code live at the main InstantHMR
-repository.
+repository: <https://github.com/mohamdev/InstantHMR>.
## Files
diff --git a/models/instanthmr.gif b/models/instanthmr.gif
new file mode 100644
index 0000000..c9345d4
Binary files /dev/null and b/models/instanthmr.gif differ