workshop: sync notebooks/berk_workshop_parallax_v1.ipynb

notebooks/berk_workshop_parallax_v1.ipynb

@@ -0,0 +1,213 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "id": "d7f1098c",
+   "metadata": {},
+   "source": [
+    "# Workshop: parallax phase retrieval on real gold (Colab T4)\n",
+    "\n",
+    "[![Open in Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/gist/bobleesj/96ea70cbb0d3f97b32d13c0e7a5448ec/berk_workshop_parallax_v1.ipynb)\n",
+    "\n",
+    "Single-shot phase retrieval via the **parallax** mode of `DirectPtychography`.\n",
+    "Same real-gold 4D-STEM data from Hugging Face. No iterative ptycho yet — just\n",
+    "one direct deconvolution that gives you a usable phase image in seconds.\n",
+    "\n",
+    "Two installs (same as the BF/DF/DPC workshop): `quantem.widget` (TestPyPI\n",
+    "prerelease) + `quantem` from the `berk-workshop` branch on `bobleesj/quantem`.\n",
+    "\n",
+    "**Total runtime: 2-3 minutes on Colab T4.**"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "27f47377",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "!pip install -q --pre --extra-index-url https://test.pypi.org/simple/ quantem.widget huggingface_hub\n",
+    "!pip install -q git+https://github.com/bobleesj/quantem.git@berk-workshop"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "a3ec89be",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import quantem as em\n",
+    "import quantem.widget\n",
+    "import torch\n",
+    "\n",
+    "# cuDNN grid_sample currently fails on the parallax shift-origin step for these\n",
+    "# dimensions; disable to use the deterministic CUDA path (no perf impact at this\n",
+    "# data size). Tracked upstream.\n",
+    "torch.backends.cudnn.enabled = False\n",
+    "\n",
+    "print(\"quantem        \", em.__version__)\n",
+    "print(\"quantem.widget \", quantem.widget.__version__)\n",
+    "print(\"torch          \", torch.__version__, \"(cuDNN disabled)\")\n",
+    "print(\"cuda available:\", torch.cuda.is_available(),\n",
+    "      torch.cuda.get_device_name(0) if torch.cuda.is_available() else \"(no GPU)\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "08f7a2a4",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import os, json\n",
+    "import numpy as np\n",
+    "from huggingface_hub import snapshot_download\n",
+    "\n",
+    "folder = snapshot_download(\"bobleesj/quantem-data\", repo_type=\"dataset\",\n",
+    "                           allow_patterns=[\"4dstem/gold_512_npy_bin8/*\"])\n",
+    "asset = os.path.join(folder, \"4dstem\", \"gold_512_npy_bin8\")\n",
+    "data = np.load(os.path.join(asset, \"data.npy\")).astype(np.float32)   # float for DirectPtychography\n",
+    "data = np.ascontiguousarray(data)\n",
+    "meta = json.load(open(os.path.join(asset, \"meta.json\")))\n",
+    "\n",
+    "dset = em.core.datastructures.Dataset4dstem.from_array(\n",
+    "    data, sampling=meta[\"sampling\"], units=meta[\"units\"], name=meta[\"name\"],\n",
+    ")\n",
+    "print(f\"dataset: shape {dset.shape}, dtype {dset.array.dtype}\")\n",
+    "print(f\"sampling {meta['sampling']} {meta['units']}\")\n",
+    "print(f\"optics: {meta['voltage_kV']} kV, probe {meta['probe_semiangle_mrad']} mrad\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "7cd2a2fc",
+   "metadata": {},
+   "source": [
+    "## Step 1 — Build the DirectPtychography object\n",
+    "\n",
+    "`DirectPtychography.from_dataset4d` runs CoM + origin fit + auto-rotation\n",
+    "estimate internally. Pass the optics in SI: energy in eV, semi-angle cutoff in\n",
+    "radians."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "8ef4082e",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from quantem.diffractive_imaging import DirectPtychography\n",
+    "\n",
+    "DEVICE = \"cuda\" if torch.cuda.is_available() else \"cpu\"\n",
+    "\n",
+    "direct = DirectPtychography.from_dataset4d(\n",
+    "    dset,\n",
+    "    energy=meta[\"voltage_kV\"] * 1e3,                          # 300 kV -> 300000 eV\n",
+    "    semiangle_cutoff=meta[\"probe_semiangle_mrad\"] * 1e-3,     # 30 mrad -> 0.030 rad\n",
+    "    rotation_angle=None,                                       # auto-estimate\n",
+    "    device=DEVICE,\n",
+    "    verbose=True,\n",
+    ")\n",
+    "print(f\"DirectPtychography built on {DEVICE}\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "a56cb2e5",
+   "metadata": {},
+   "source": [
+    "## Step 2 — Reconstruct with the parallax kernel\n",
+    "\n",
+    "`deconvolution_kernel=\"parallax\"` runs the parallax-tilt approximation (a\n",
+    "weighted shift-and-add over virtual BF images). Fast — single forward pass,\n",
+    "no iterative loop."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "db340059",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "direct.reconstruct(\n",
+    "    deconvolution_kernel=\"parallax\",\n",
+    "    parallax_flip_phase=False,\n",
+    "    verbose=True,\n",
+    ")\n",
+    "print(\"parallax recon done\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "ef1cc477",
+   "metadata": {},
+   "source": [
+    "## Step 3 — Visualize the recovered phase\n",
+    "\n",
+    "The parallax phase image lives on the GPU; pull to CPU once for the widget."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "feb7fbfe",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "phase = direct.corrected_bf                       # (scan_r, scan_c) float32 on cuda\n",
+    "print(f\"phase shape {tuple(phase.shape)}, device {phase.device}\")\n",
+    "print(f\"phase range [{phase.min().item():.3f}, {phase.max().item():.3f}]\")\n",
+    "\n",
+    "quantem.widget.Show2D(\n",
+    "    phase.detach().cpu().numpy(),\n",
+    "    title=\"Parallax phase (corrected_bf)\",\n",
+    "    sampling=meta[\"sampling\"][:2],\n",
+    "    units=meta[\"units\"][:2],\n",
+    "    cmap=\"gray\",\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "478edc58",
+   "metadata": {},
+   "source": [
+    "## What you just did\n",
+    "\n",
+    "1. Loaded real gold 4D-STEM from Hugging Face → numpy-backed `Dataset4dstem`.\n",
+    "2. Built `DirectPtychography` with the gold's actual optics (300 kV, 30 mrad).\n",
+    "3. Ran a single-shot parallax reconstruction (no iterative loop).\n",
+    "4. Rendered `direct.corrected_bf` as an interactive `Show2D` widget.\n",
+    "\n",
+    "Single forward pass on the T4. Companion notebook `berk_workshop_v1.ipynb`\n",
+    "covers browse + BF/DF + DPC on the same data.\n",
+    "\n",
+    "## Try next\n",
+    "\n",
+    "- `parallax_flip_phase=True` if the phase comes out inverted.\n",
+    "- Swap to `gold_512_npy_bin4` for a finer (4× larger) detector.\n",
+    "- v2 will add iterative ptychography (`PtychoLite`) for higher-resolution phase."
+   ]
+  }
+ ],
+ "metadata": {
+  "accelerator": "GPU",
+  "colab": {
+   "gpuType": "T4",
+   "include_colab_link": true,
+   "provenance": []
+  },
+  "kernelspec": {
+   "display_name": "Python 3",
+   "name": "python3"
+  },
+  "language_info": {
+   "name": "python"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}