Develop

.gitignore

@@ -11,4 +11,5 @@ checkpoints/*.pt
 demo/backend/checkpoints/*.pt
 datasets/*
 *.zip
-*.txt
+*.txt
+data

convert.ipynb

@@ -0,0 +1,179 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "id": "84f466df",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Converting sequence: bike-packing\n",
+      "→ Saved gray masks in data/sam2_preds_og_gray/bike-packing\n",
+      "Converting sequence: blackswan\n",
+      "→ Saved gray masks in data/sam2_preds_og_gray/blackswan\n",
+      "Converting sequence: bmx-trees\n",
+      "→ Saved gray masks in data/sam2_preds_og_gray/bmx-trees\n",
+      "Converting sequence: breakdance\n",
+      "→ Saved gray masks in data/sam2_preds_og_gray/breakdance\n",
+      "Converting sequence: camel\n",
+      "→ Saved gray masks in data/sam2_preds_og_gray/camel\n",
+      "Converting sequence: car-roundabout\n",
+      "→ Saved gray masks in data/sam2_preds_og_gray/car-roundabout\n",
+      "Converting sequence: car-shadow\n",
+      "→ Saved gray masks in data/sam2_preds_og_gray/car-shadow\n",
+      "Converting sequence: cows\n",
+      "→ Saved gray masks in data/sam2_preds_og_gray/cows\n",
+      "Converting sequence: dance-twirl\n",
+      "→ Saved gray masks in data/sam2_preds_og_gray/dance-twirl\n",
+      "Converting sequence: dog\n",
+      "→ Saved gray masks in data/sam2_preds_og_gray/dog\n",
+      "Converting sequence: dogs-jump\n",
+      "→ Saved gray masks in data/sam2_preds_og_gray/dogs-jump\n",
+      "Converting sequence: drift-chicane\n",
+      "→ Saved gray masks in data/sam2_preds_og_gray/drift-chicane\n",
+      "Converting sequence: drift-straight\n",
+      "→ Saved gray masks in data/sam2_preds_og_gray/drift-straight\n",
+      "Converting sequence: goat\n",
+      "→ Saved gray masks in data/sam2_preds_og_gray/goat\n",
+      "Converting sequence: gold-fish\n",
+      "→ Saved gray masks in data/sam2_preds_og_gray/gold-fish\n",
+      "Converting sequence: horsejump-high\n",
+      "→ Saved gray masks in data/sam2_preds_og_gray/horsejump-high\n",
+      "Converting sequence: india\n",
+      "→ Saved gray masks in data/sam2_preds_og_gray/india\n",
+      "Converting sequence: judo\n",
+      "→ Saved gray masks in data/sam2_preds_og_gray/judo\n",
+      "Converting sequence: kite-surf\n",
+      "→ Saved gray masks in data/sam2_preds_og_gray/kite-surf\n",
+      "Converting sequence: lab-coat\n",
+      "→ Saved gray masks in data/sam2_preds_og_gray/lab-coat\n",
+      "Converting sequence: libby\n",
+      "→ Saved gray masks in data/sam2_preds_og_gray/libby\n",
+      "Converting sequence: loading\n",
+      "→ Saved gray masks in data/sam2_preds_og_gray/loading\n",
+      "Converting sequence: mbike-trick\n",
+      "→ Saved gray masks in data/sam2_preds_og_gray/mbike-trick\n",
+      "Converting sequence: motocross-jump\n",
+      "→ Saved gray masks in data/sam2_preds_og_gray/motocross-jump\n",
+      "Converting sequence: paragliding-launch\n",
+      "→ Saved gray masks in data/sam2_preds_og_gray/paragliding-launch\n",
+      "Converting sequence: parkour\n",
+      "→ Saved gray masks in data/sam2_preds_og_gray/parkour\n",
+      "Converting sequence: pigs\n",
+      "→ Saved gray masks in data/sam2_preds_og_gray/pigs\n",
+      "Converting sequence: scooter-black\n",
+      "→ Saved gray masks in data/sam2_preds_og_gray/scooter-black\n",
+      "Converting sequence: shooting\n",
+      "→ Saved gray masks in data/sam2_preds_og_gray/shooting\n",
+      "Converting sequence: soapbox\n",
+      "→ Saved gray masks in data/sam2_preds_og_gray/soapbox\n",
+      "All sequences converted. Now run DAVIS evaluation on data/sam2_preds_og_gray\n"
+     ]
+    }
+   ],
+   "source": [
+    "import os\n",
+    "from pathlib import Path\n",
+    "import imageio.v3 as iio\n",
+    "import numpy as np\n",
+    "\n",
+    "# ── USER CONFIG ─────────────────────────────────────────────────────────\n",
+    "SOURCE_DIR = Path(\"./data/sam2_preds_og\")  # your colored‐RGB predictions\n",
+    "TARGET_DIR = Path(\"./data/sam2_preds_og_gray\")   # where we'll save single‐channel PNGs\n",
+    "\n",
+    "TARGET_DIR.mkdir(parents=True, exist_ok=True)\n",
+    "\n",
+    "# ── HELPER: find the “object‐colors → ID” mapping from frame 0 ─────────────\n",
+    "def extract_color_to_id_map(first_color_png):\n",
+    "    \"\"\"\n",
+    "    Read the RGB image first_color_png (H×W×3), find all non‐black colors,\n",
+    "    and assign them IDs 1,2,3,… in sorted order. Returns:\n",
+    "      • color2id: dict mapping (R,G,B) tuples → integer ID\n",
+    "      • H, W: image dimensions\n",
+    "    \"\"\"\n",
+    "    rgb = iio.imread(str(first_color_png))\n",
+    "    if rgb.ndim != 3 or rgb.shape[2] != 3:\n",
+    "        raise RuntimeError(f\"Expected a (H, W, 3) image at {first_color_png}\")\n",
+    "\n",
+    "    flat = rgb.reshape(-1, 3)\n",
+    "    uniq = np.unique(flat, axis=0)       # (K,3) array of all colors present\n",
+    "    # Drop black (0,0,0):\n",
+    "    non_black = [tuple(c) for c in uniq if not np.all(c == 0)]\n",
+    "    if len(non_black) == 0:\n",
+    "        raise RuntimeError(f\"No non‐black colors found in {first_color_png}\")\n",
+    "\n",
+    "    # Sort by RGB lex order (optional) to assign stable IDs:\n",
+    "    non_black.sort()\n",
+    "    color2id = {color: (i + 1) for i, color in enumerate(non_black)}\n",
+    "    return color2id, rgb.shape[0], rgb.shape[1]\n",
+    "\n",
+    "# ── MAIN LOOP: for each sequence, convert every frame’s RGB to single‐channel IDs ───\n",
+    "for seq_folder in sorted(SOURCE_DIR.iterdir()):\n",
+    "    if not seq_folder.is_dir():\n",
+    "        continue\n",
+    "\n",
+    "    print(f\"Converting sequence: {seq_folder.name}\")\n",
+    "    out_seq = TARGET_DIR / seq_folder.name\n",
+    "    out_seq.mkdir(parents=True, exist_ok=True)\n",
+    "\n",
+    "    # (1) find frame 00000.png and build the color→ID lookup\n",
+    "    first_frame = seq_folder / \"00000.png\"\n",
+    "    if not first_frame.exists():\n",
+    "        raise RuntimeError(f\"Cannot find {first_frame}\")\n",
+    "\n",
+    "    color2id, H, W = extract_color_to_id_map(first_frame)\n",
+    "    # Example: color2id might be {(200,0,0): 1, (0,200,0): 2}\n",
+    "\n",
+    "    # (2) iterate over all PNGs in this sequence folder (00000.png, 00001.png, …)\n",
+    "    all_frames = sorted(seq_folder.glob(\"*.png\"))\n",
+    "\n",
+    "    for frame_path in all_frames:\n",
+    "        rgb = iio.imread(str(frame_path))\n",
+    "        if rgb.ndim != 3 or rgb.shape[:2] != (H, W):\n",
+    "            raise RuntimeError(f\"Unexpected image shape in {frame_path}: {rgb.shape}\")\n",
+    "\n",
+    "        # Build a blank H×W array of uint8\n",
+    "        id_map = np.zeros((H, W), dtype=np.uint8)\n",
+    "\n",
+    "        # For each distinct color in color2id, mask and assign ID\n",
+    "        # (Pixels that remain black → ID=0)\n",
+    "        for (R, G, B), obj_id in color2id.items():\n",
+    "            mask = (rgb[:, :, 0] == R) & (rgb[:, :, 1] == G) & (rgb[:, :, 2] == B)\n",
+    "            if mask.any():\n",
+    "                id_map[mask] = obj_id\n",
+    "\n",
+    "        # Save the new single‐channel PNG\n",
+    "        out_path = out_seq / frame_path.name\n",
+    "        iio.imwrite(str(out_path), id_map)\n",
+    "\n",
+    "    print(f\"→ Saved gray masks in {out_seq}\")\n",
+    "\n",
+    "print(\"All sequences converted. Now run DAVIS evaluation on\", TARGET_DIR)\n"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "sam2",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.13.2"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}