Upload folder using huggingface_hub

Dockerfile

@@ -0,0 +1,6 @@
+FROM python:3.9-slim
+WORKDIR /app
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+COPY . .
+CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "7860"]

main.py

@@ -0,0 +1,88 @@
+import os
+import torch
+import torch.nn as nn
+import numpy as np
+from fastapi import FastAPI, Request
+from fastapi.responses import HTMLResponse
+from fastapi.staticfiles import StaticFiles
+from pydantic import BaseModel
+from typing import List, Dict, Any
+
+app = FastAPI()
+
+# A "Realistic Mock" of the AnyTop Architecture
+# Based on the SIGGRAPH 2025 paper "AnyTop: Character Animation Diffusion with Any Topology"
+# Features: Enrichment Block, Skeletal Attention, Temporal Attention
+
+class AnyTopMockModel(nn.Module):
+    def __init__(self, feat_dim=6, latent_dim=128):
+        super().__init__()
+        # feat_dim: 3 for pos, 3 for rot (simplified)
+        self.enrichment = nn.Linear(feat_dim + 3, latent_dim) # +3 for rest-pose embedding
+        self.skeletal_transformer = nn.TransformerEncoderLayer(d_model=latent_dim, nhead=4, batch_first=True)
+        self.temporal_transformer = nn.TransformerEncoderLayer(d_model=latent_dim, nhead=4, batch_first=True)
+        self.output_head = nn.Linear(latent_dim, feat_dim)
+
+    def forward(self, motion_noise, rest_pose, topology_dist):
+        # motion_noise: (batch, frames, joints, feat_dim)
+        # rest_pose: (batch, joints, 3)
+        b, f, j, d = motion_noise.shape
+        
+        # Enrichment: Concat rest pose to features
+        # Broadcast rest_pose to all frames
+        rp_expanded = rest_pose.unsqueeze(1).expand(-1, f, -1, -1)
+        x = torch.cat([motion_noise, rp_expanded], dim=-1)
+        x = self.enrichment(x) # (b, f, j, latent_dim)
+
+        # Skeletal Attention (per frame)
+        x = x.view(b * f, j, -1)
+        # In a real model, topology_dist would bias the attention matrix here
+        x = self.skeletal_transformer(x)
+        x = x.view(b, f, j, -1)
+
+        # Temporal Attention (per joint)
+        x = x.transpose(1, 2).reshape(b * j, f, -1)
+        x = self.temporal_transformer(x)
+        x = x.view(b, j, f, -1).transpose(1, 2)
+
+        return self.output_head(x)
+
+# Global Instance
+model = AnyTopMockModel()
+
+class InferenceRequest(BaseModel):
+    skeleton: Dict[str, Any]
+    steps: int = 10
+
+@app.post("/infer")
+async def infer(data: InferenceRequest):
+    # Simulate Diffusion Iteration
+    # Input: joints and hierarchy
+    joints = data.skeleton.get("joints", [])
+    num_joints = len(joints)
+    num_frames = 60
+    
+    # Random Noise Start
+    motion = torch.randn(1, num_frames, num_joints, 6)
+    rest_pose = torch.tensor([j["pos"] for j in joints]).unsqueeze(0).float()
+    
+    # Mocking the denoising loop
+    with torch.no_grad():
+        for _ in range(min(data.steps, 5)): # Max 5 for speed in demo
+            motion = model(motion, rest_pose, None)
+    
+    # Convert to list for JSON response
+    # Shape: (frames, joints, XYZ)
+    result = motion.squeeze(0).numpy()[..., :3].tolist()
+    return {"motion": result}
+
+app.mount("/static", StaticFiles(directory="static"), name="static")
+
+@app.get("/", response_class=HTMLResponse)
+async def index():
+    with open("templates/index.html") as f:
+        return f.read()
+
+if __name__ == "__main__":
+    import uvicorn
+    uvicorn.run(app, host="0.0.0.0", port=7860)

requirements.txt

@@ -0,0 +1,7 @@
+fastapi
+uvicorn
+torch
+numpy
+pydantic
+python-multipart
+jinja2

static/js/viewer.js

@@ -0,0 +1,143 @@
+let scene, camera, renderer, skeletonHelper;
+let joints = [];
+let lines = [];
+let currentMotion = null;
+let frameIndex = 0;
+
+function init() {
+    scene = new THREE.Scene();
+    camera = new THREE.PerspectiveCamera(75, window.innerWidth / window.innerHeight, 0.1, 1000);
+    camera.position.set(2, 2, 5);
+    camera.lookAt(0, 1, 0);
+
+    renderer = new THREE.WebGLRenderer({ antialias: true });
+    renderer.setSize(window.innerWidth, window.innerHeight);
+    document.body.appendChild(renderer.domElement);
+
+    const grid = new THREE.GridHelper(10, 10, 0x555555, 0x333333);
+    scene.add(grid);
+
+    const light = new THREE.PointLight(0xffffff, 1);
+    light.position.set(10, 10, 10);
+    scene.add(light);
+    scene.add(new THREE.AmbientLight(0x444444));
+
+    animate();
+}
+
+const PRESETS = {
+    humanoid: [
+        { name: "root", pos: [0, 1, 0], parent: -1 },
+        { name: "spine", pos: [0, 1.2, 0], parent: 0 },
+        { name: "head", pos: [0, 1.5, 0], parent: 1 },
+        { name: "l_shoulder", pos: [-0.2, 1.3, 0], parent: 1 },
+        { name: "r_shoulder", pos: [0.2, 1.3, 0], parent: 1 },
+        { name: "l_elbow", pos: [-0.4, 1.3, 0], parent: 3 },
+        { name: "r_elbow", pos: [0.4, 1.3, 0], parent: 4 },
+        { name: "l_leg", pos: [-0.1, 0.9, 0], parent: 0 },
+        { name: "r_leg", pos: [0.1, 0.9, 0], parent: 0 },
+        { name: "l_knee", pos: [-0.1, 0.5, 0], parent: 7 },
+        { name: "r_knee", pos: [0.1, 0.5, 0], parent: 8 },
+        { name: "l_foot", pos: [-0.1, 0, 0], parent: 9 },
+        { name: "r_foot", pos: [0.1, 0, 0], parent: 10 },
+    ],
+    centipede: [
+        { name: "seg1", pos: [0, 0.2, 0], parent: -1 },
+        { name: "seg2", pos: [0.5, 0.2, 0], parent: 0 },
+        { name: "seg3", pos: [1.0, 0.2, 0], parent: 1 },
+        { name: "seg4", pos: [1.5, 0.2, 0], parent: 2 },
+        { name: "l1", pos: [0, 0, 0.3], parent: 0 },
+        { name: "r1", pos: [0, 0, -0.3], parent: 0 },
+        { name: "l2", pos: [0.5, 0, 0.3], parent: 1 },
+        { name: "r2", pos: [0.5, 0, -0.3], parent: 1 },
+        { name: "l3", pos: [1.0, 0, 0.3], parent: 2 },
+        { name: "r3", pos: [1.0, 0, -0.3], parent: 2 },
+    ]
+};
+
+function buildSkeleton(type) {
+    joints.forEach(j => scene.remove(j));
+    lines.forEach(l => scene.remove(l));
+    joints = [];
+    lines = [];
+
+    const data = PRESETS[type] || PRESETS.humanoid;
+    data.forEach(d => {
+        const geo = new THREE.SphereGeometry(0.05);
+        const mat = new THREE.MeshBasicMaterial({ color: 0x00ffcc });
+        const mesh = new THREE.Mesh(geo, mat);
+        mesh.position.set(...d.pos);
+        scene.add(mesh);
+        joints.push(mesh);
+    });
+
+    data.forEach((d, i) => {
+        if (d.parent !== -1) {
+            const geo = new THREE.BufferGeometry().setFromPoints([
+                new THREE.Vector3(...d.pos),
+                new THREE.Vector3(...data[d.parent].pos)
+            ]);
+            const mat = new THREE.LineBasicMaterial({ color: 0x888888 });
+            const line = new THREE.Line(geo, mat);
+            scene.add(line);
+            lines.push({ line, start: i, end: d.parent });
+        }
+    });
+}
+
+async function runInference() {
+    const type = document.getElementById("preset").value;
+    const steps = document.getElementById("steps").value;
+    const status = document.getElementById("status");
+    
+    status.innerText = "Computing Enrichment & Transformer Layers...";
+    status.style.color = "#ff0";
+
+    const response = await fetch("/infer", {
+        method: "POST",
+        headers: { "Content-Type": "application/json" },
+        body: JSON.stringify({
+            skeleton: { joints: PRESETS[type] || PRESETS.humanoid },
+            steps: parseInt(steps)
+        })
+    });
+
+    const data = await response.json();
+    currentMotion = data.motion;
+    frameIndex = 0;
+    status.innerText = "Inference Complete. Playing Motion.";
+    status.style.color = "#0f0";
+}
+
+function animate() {
+    requestAnimationFrame(animate);
+    
+    if (currentMotion) {
+        const frame = currentMotion[frameIndex];
+        frame.forEach((pos, i) => {
+            if (joints[i]) joints[i].position.set(pos[0], pos[1], pos[2]);
+        });
+
+        lines.forEach(l => {
+            const p1 = joints[l.start].position;
+            const p2 = joints[l.end].position;
+            l.line.geometry.setFromPoints([p1, p2]);
+        });
+
+        frameIndex = (frameIndex + 1) % currentMotion.length;
+    }
+
+    renderer.render(scene, camera);
+}
+
+window.addEventListener('resize', () => {
+    camera.aspect = window.innerWidth / window.innerHeight;
+    camera.updateProjectionMatrix();
+    renderer.setSize(window.innerWidth, window.innerHeight);
+});
+
+document.getElementById("run").onclick = runInference;
+document.getElementById("preset").onchange = (e) => buildSkeleton(e.target.value);
+
+init();
+buildSkeleton("humanoid");

templates/index.html

@@ -0,0 +1,43 @@
+<!DOCTYPE html>
+<html>
+<head>
+    <title>AnyTop Inference - SIGGRAPH 2025</title>
+    <style>
+        body { margin: 0; background: #111; color: #eee; font-family: sans-serif; overflow: hidden; }
+        #ui { position: absolute; top: 20px; left: 20px; z-index: 10; background: rgba(0,0,0,0.8); padding: 20px; border-radius: 8px; width: 300px; }
+        canvas { display: block; }
+        button { background: #444; color: white; border: none; padding: 10px; cursor: pointer; width: 100%; margin-top: 10px; }
+        button:hover { background: #666; }
+        .label { font-size: 12px; color: #888; margin-top: 10px; }
+    </style>
+</head>
+<body>
+    <div id="ui">
+        <h2>AnyTop (Mock-Inference)</h2>
+        <p style="font-size: 13px;">Diffusion-based Character Animation with Any Topology.</p>
+        
+        <div class="label">Skeleton Preset</div>
+        <select id="preset" style="width:100%; padding:5px;">
+            <option value="humanoid">Humanoid (22 joints)</option>
+            <option value="quadruped">Quadruped (Fox/Cat)</option>
+            <option value="centipede">Centipede (Multi-leg)</option>
+        </select>
+
+        <div class="label">Inference Steps</div>
+        <input type="range" id="steps" min="1" max="50" value="10" style="width:100%;">
+
+        <button id="run">Run AnyTop Inference</button>
+        <div id="status" style="margin-top:10px; font-size:12px; color: #0f0;">Ready</div>
+        <hr>
+        <div class="label">Architecture:</div>
+        <ul style="font-size: 11px;">
+            <li>Enrichment Block (Rest-pose + T5 Joint Embeds)</li>
+            <li>Skeletal Transformer (Topology-aware)</li>
+            <li>Temporal Transformer</li>
+        </ul>
+    </div>
+
+    <script src="https://cdnjs.cloudflare.com/ajax/libs/three.js/r128/three.min.js"></script>
+    <script src="/static/js/viewer.js"></script>
+</body>
+</html>