workshop: BF+DF single widget; Show4DSTEM width; drop cross-modality

notebooks/berk_workshop_v1.ipynb

@@ -175,7 +175,9 @@
     "    torch.from_numpy(data).to(\"cuda\" if torch.cuda.is_available() else \"cpu\"),\n",
     "    sampling=meta[\"sampling\"], units=meta[\"units\"], name=meta[\"name\"],\n",
     ")\n",
-    "quantem.widget.Show4DSTEM(dset_torch)"
+    "w = quantem.widget.Show4DSTEM(dset_torch)\n",
+    "w.layout.width = \"1000px\"      # ipywidgets layout — best-effort; widget JS may override\n",
+    "w"
    ]
   },
   {
@@ -183,11 +185,11 @@
    "id": "c1faa1b5",
    "metadata": {},
    "source": [
-    "## Step 2 — Bright field (BF) via `get_virtual_image`\n",
+    "## Step 2 — Bright field + Dark field, side by side (independent contrast)\n",
     "\n",
-    "Upstream `Dataset4dstem.get_virtual_image(mode=\"circle\", geometry=((cy, cx), r))`\n",
-    "makes the aperture mask + per-scan-position sum + returns a `Dataset2d`. One\n",
-    "line. (`Show2D` shown alone so contrast is not yoked to anything else.)"
+    "`Dataset4dstem.get_virtual_image(mode=\"circle\"|\"annular\", ...)` makes the\n",
+    "aperture masks + per-scan-position sum. One `Show2D` widget with both panels;\n",
+    "`link_contrast=False` so each panel scales to its own min/max."
    ]
   },
   {
@@ -220,78 +222,24 @@
     }
    ],
    "source": [
+    "import math\n",
+    "\n",
     "H, W = dset.shape[-2:]\n",
     "cy, cx = H / 2, W / 2                                    # hardcoded geometric center\n",
     "BF_RADIUS_PX = 6.0\n",
+    "R_MAX = math.hypot(cy, cx)                                # corner-of-detector radius\n",
     "\n",
-    "bf_ds = dset.get_virtual_image(\n",
-    "    mode=\"circle\",\n",
-    "    geometry=((cy, cx), BF_RADIUS_PX),\n",
-    "    name=\"BF\",\n",
-    ")\n",
+    "bf_ds = dset.get_virtual_image(mode=\"circle\",  geometry=((cy, cx), BF_RADIUS_PX), name=\"BF\")\n",
+    "df_ds = dset.get_virtual_image(mode=\"annular\", geometry=((cy, cx), (BF_RADIUS_PX, R_MAX)), name=\"DF\")\n",
     "print(f\"BF range [{bf_ds.array.min():.1f}, {bf_ds.array.max():.1f}]\")\n",
-    "\n",
-    "quantem.widget.Show2D(\n",
-    "    bf_ds.array, title=\"Bright field\",\n",
-    "    sampling=meta[\"sampling\"][:2], units=meta[\"units\"][:2],\n",
-    "    cmap=\"gray\",\n",
-    ")"
-   ]
-  },
-  {
-   "cell_type": "markdown",
-   "id": "6cc22c3c",
-   "metadata": {},
-   "source": [
-    "## Step 3 — Dark field (DF) via `get_virtual_image` (annular)\n",
-    "\n",
-    "Annular mask: everything between the BF radius and the detector edge."
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 43,
-   "id": "79e0c9fc",
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "DF range [100550.0, 181611.0]\n"
-     ]
-    },
-    {
-     "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "e3316a8cbb794108b8bea37e1c9e7707",
-       "version_major": 2,
-       "version_minor": 1
-      },
-      "text/plain": [
-       "Show2D(512×512, cmap=gray)"
-      ]
-     },
-     "execution_count": 43,
-     "metadata": {},
-     "output_type": "execute_result"
-    }
-   ],
-   "source": [
-    "import math\n",
-    "R_MAX = math.hypot(cy, cx)   # corner-of-detector radius\n",
-    "\n",
-    "df_ds = dset.get_virtual_image(\n",
-    "    mode=\"annular\",\n",
-    "    geometry=((cy, cx), (BF_RADIUS_PX, R_MAX)),\n",
-    "    name=\"DF\",\n",
-    ")\n",
     "print(f\"DF range [{df_ds.array.min():.1f}, {df_ds.array.max():.1f}]\")\n",
     "\n",
     "quantem.widget.Show2D(\n",
-    "    df_ds.array, title=\"Dark field\",\n",
+    "    [bf_ds.array, df_ds.array],\n",
+    "    labels=[\"Bright field\", \"Dark field\"],\n",
     "    sampling=meta[\"sampling\"][:2], units=meta[\"units\"][:2],\n",
     "    cmap=\"gray\",\n",
+    "    link_contrast=False,\n",
     ")"
    ]
   },
@@ -300,7 +248,7 @@
    "id": "50bcd0fe",
    "metadata": {},
    "source": [
-    "## Step 4 — DPC via `CenterOfMassOriginModel`\n",
+    "## Step 3 — DPC via `CenterOfMassOriginModel`\n",
     "\n",
     "Per-scan-position centroid (CoM) on the GPU. Two-cell call, two interactive panels."
    ]
@@ -356,7 +304,7 @@
    "id": "e8050ac2",
    "metadata": {},
    "source": [
-    "## Step 4b — Scan-detector rotation from CoM curl\n",
+    "## Step 3b — Scan-detector rotation from CoM curl\n",
     "\n",
     "`DirectPtychography` needs the rotation between the scan axes and the detector\n",
     "axes. Find it by sweeping rotation angles and minimizing the curl of the\n",
@@ -407,7 +355,7 @@
    "id": "6c89c61d",
    "metadata": {},
    "source": [
-    "## Step 5 — Phase retrieval: `DirectPtychography` + optuna aberration fit\n",
+    "## Step 4 — Phase retrieval: `DirectPtychography` + optuna aberration fit\n",
     "\n",
     "Build `DirectPtychography` with the **rotation found in Step 4b** (`best_rotation_rad`),\n",
     "then **fit C10, C12, phi12 from the data** with a 30-trial optuna loop (~5 s on T4).\n",
@@ -504,7 +452,7 @@
    "id": "5ffc4fd8",
    "metadata": {},
    "source": [
-    "## Step 6 — All three kernels side by side\n",
+    "## Step 5 — All three kernels side by side\n",
     "\n",
     "`link_contrast=False` so every kernel gets its own min/max (the SSB output is\n",
     "~3 orders of magnitude smaller than ICOM)."
@@ -542,153 +490,6 @@
     ")"
    ]
   },
-  {
-   "cell_type": "markdown",
-   "id": "c2d0e9b4",
-   "metadata": {},
-   "source": [
-    "## Step 7 — Phase retrieval vs classic imaging\n",
-    "\n",
-    "The workshop punchline.\n",
-    "\n",
-    "- **BF / DF**: intensity contrast from inside / outside the BF disk. Limited\n",
-    "  by probe size; atomic-lattice fringes mostly washed out.\n",
-    "- **ICOM, parallax, SSB**: deconvolution of the full CBED against the probe\n",
-    "  transfer function. Sharper contrast at the same dose.\n",
-    "\n",
-    "Each panel its own `Show2D` widget — independent contrast."
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 48,
-   "id": "5367fee4",
-   "metadata": {},
-   "outputs": [
-    {
-     "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "440f0bb7f30640178f0b0e0112dd3858",
-       "version_major": 2,
-       "version_minor": 1
-      },
-      "text/plain": [
-       "Show2D(512×512, cmap=gray)"
-      ]
-     },
-     "execution_count": 48,
-     "metadata": {},
-     "output_type": "execute_result"
-    }
-   ],
-   "source": [
-    "quantem.widget.Show2D(bf_ds.array, title=\"BF — intensity inside disk\", sampling=meta[\"sampling\"][:2], units=meta[\"units\"][:2], cmap=\"gray\")"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 49,
-   "id": "6058a9b6",
-   "metadata": {},
-   "outputs": [
-    {
-     "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "985bbf76e0494a729d899da1b20e50b7",
-       "version_major": 2,
-       "version_minor": 1
-      },
-      "text/plain": [
-       "Show2D(512×512, cmap=gray)"
-      ]
-     },
-     "execution_count": 49,
-     "metadata": {},
-     "output_type": "execute_result"
-    }
-   ],
-   "source": [
-    "quantem.widget.Show2D(df_ds.array, title=\"DF — intensity outside disk\", sampling=meta[\"sampling\"][:2], units=meta[\"units\"][:2], cmap=\"gray\")"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 50,
-   "id": "faf40b28",
-   "metadata": {},
-   "outputs": [
-    {
-     "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "522b3f5c064a4549b637db985960c9b7",
-       "version_major": 2,
-       "version_minor": 1
-      },
-      "text/plain": [
-       "Show2D(512×512, cmap=magma)"
-      ]
-     },
-     "execution_count": 50,
-     "metadata": {},
-     "output_type": "execute_result"
-    }
-   ],
-   "source": [
-    "quantem.widget.Show2D(phases[\"icom\"], title=\"ICOM — first-moment phase\", sampling=meta[\"sampling\"][:2], units=meta[\"units\"][:2], cmap=\"gray\")"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 51,
-   "id": "63197495",
-   "metadata": {},
-   "outputs": [
-    {
-     "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "2970116ba0134cd88d32e9adfa3c3a43",
-       "version_major": 2,
-       "version_minor": 1
-      },
-      "text/plain": [
-       "Show2D(512×512, cmap=gray)"
-      ]
-     },
-     "execution_count": 51,
-     "metadata": {},
-     "output_type": "execute_result"
-    }
-   ],
-   "source": [
-    "quantem.widget.Show2D(phases[\"parallax\"], title=\"parallax — phase retrieval\", sampling=meta[\"sampling\"][:2], units=meta[\"units\"][:2], cmap=\"gray\")"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 52,
-   "id": "52f7726c",
-   "metadata": {},
-   "outputs": [
-    {
-     "data": {
-      "application/vnd.jupyter.widget-view+json": {
-       "model_id": "7f9e0ed1104b43a58814e76ae234bd22",
-       "version_major": 2,
-       "version_minor": 1
-      },
-      "text/plain": [
-       "Show2D(512×512, cmap=gray)"
-      ]
-     },
-     "execution_count": 52,
-     "metadata": {},
-     "output_type": "execute_result"
-    }
-   ],
-   "source": [
-    "quantem.widget.Show2D(phases[\"ssb\"], title=\"SSB — phase retrieval\", sampling=meta[\"sampling\"][:2], units=meta[\"units\"][:2], cmap=\"gray\")"
-   ]
-  },
   {
    "cell_type": "markdown",
    "id": "33a873d6",
@@ -697,11 +498,11 @@
     "## What you just did\n",
     "\n",
     "1. Loaded real 4D-STEM gold from Hugging Face → torch GPU + numpy `Dataset4dstem`.\n",
-    "2. Browsed it with `Show4DSTEM`.\n",
-    "3. BF, DF: upstream `Dataset4dstem.get_virtual_image(mode=\"circle\"|\"annular\", ...)`, separate `Show2D` widgets.\n",
-    "4. DPC: upstream `CenterOfMassOriginModel.from_dataset(..., device=\"cuda\").calculate_origin()`.\n",
-    "5. Built `DirectPtychography`; ran a 30-trial optuna fit of C10/C12/phi12; swept three deconvolution kernels (ICOM, parallax, SSB).\n",
-    "6. Compared all five modalities — each in its own widget.\n",
+    "2. Browsed with `Show4DSTEM`.\n",
+    "3. BF + DF in one `Show2D` widget (independent contrast) via `Dataset4dstem.get_virtual_image`.\n",
+    "4. DPC via upstream `CenterOfMassOriginModel.calculate_origin()`.\n",
+    "5. Found scan-detector rotation by minimizing CoM curl (Step 3b).\n",
+    "6. Built `DirectPtychography` with that rotation; ran a 30-trial optuna fit of C10/C12/phi12; swept ICOM / parallax / SSB.\n",
     "\n",
     "| Method | What it uses |\n",
     "|---|---|\n",
@@ -710,13 +511,13 @@
     "| ICOM, parallax, SSB | full CBED at every scan position; aberration-corrected |\n",
     "\n",
     "Takeaway: phase retrieval (especially SSB once aberrations are fit) recovers\n",
-    "contrast + resolution that BF/DF can not physically access at the same dose.\n",
+    "contrast + resolution that BF/DF can not access at the same dose.\n",
     "\n",
     "## Try next\n",
     "\n",
     "- Swap to `gold_512_npy_bin4` for a 4× finer detector.\n",
-    "- Increase optuna trials past 30 if the aberration loss isn't converging.\n",
-    "- v2 will add iterative ptychography (`PtychoLite`) for the highest-resolution phase."
+    "- Increase optuna trials past 30 if aberration loss isn't converging.\n",
+    "- v2 will add iterative ptycho (`PtychoLite`) for the highest-resolution phase."
    ]
   }
  ],