keep developing the python method

src/main/java/ai/nets/samj/models/PythonMethods.java

@@ -102,6 +102,20 @@ public class PythonMethods {
 
 
 	protected static String SAM_EVERYTHING = ""
+			+ "def calculate_pairs(masks):\n"
+			+ "    added_masks = masks.sum(2)\n"
+			+ "    inds = np.where(added_masks > 1)\n"
+			+ "    pairs = np.zeros((0, 2))\n"
+			+ "    for ii in range(inds.shape[0]):\n"
+			+ "        overlapping = np.where(masks[inds[0][ii], inds[1][ii]])[0]\n"
+			+ "        for i in range(overlapping[0].shape[0]):\n"
+			+ "            for j in range(i + 1, overlapping[0].shape[0]):\n"
+			+ "                pp = np.unique(np.array([overlapping[i], overlapping[j]])).reshape(-1, 2)\n"
+			+ "                matches = np.all(pairs == pp, axis=1)\n"
+			+ "                if not matches.any():\n"
+			+ "                    pairs = np.concatenate((pairs, pp), axis=0)\n"
+			+ "    return pairs\n"
+			+ "\n"
 			+ "def sam_everything(point_list, return_all=False):\n"
 			+ "    masks = np.zeros((input_h, input_w, 0), dtype='uint8')\n"
 			+ "    scores = []\n"
@@ -125,24 +139,28 @@ public class PythonMethods {
 			+ "        if predicted_iou[0] > 1:\n"
 			+ "            masks = np.concatenate((masks, mask.reshape(mask.shape[0], mask.shape[1], 1)), axis=1)\n"
 			+ "            scores.append(predicted_iou[0].cpu().detach().numpy())\n"
-			+ "    added_masks = masks.sum(2)\n"
-			+ "    inds = np.where(added_masks > 1)\n"
+			+ "\n"
 			+ "    # TODO do we support detection of objects within objects?\n"
-			+ "    while inds[0].shape[0] > 0:\n"
-			+ "        overlapping = np.where(masks[inds[0][0], inds[1][0]])[0]\n"
-			+ "        mask_sum = (masks[:, :, overlapping[0]] + masks[:, :, overlapping[1]])\n"
+			+ "    pairs = calculate_pairs(masks)\n"
+			+ "    while pairs.shape[0] != 0:\n"
+			+ "        pp = pairs[0]\n"
+			+ "        mask_sum = (masks[:, :, pp[0]] + masks[:, :, pp[1]])\n"
 			+ "        union = (mask_sum > 0).sum()\n"
 			+ "        intersec = (mask_sum == 2).sum()\n"
 			+ "        if intersec / union > 0.8:\n"
 			+ "            new_mask = (mask_sum > 0) * 1\n"
-			+ "            masks = np.delete(masks, overlapping[:2], axis=2)\n"
+			+ "            masks = np.delete(masks, interest, axis=2)\n"
 			+ "            masks = np.concatenate((masks, new_mask.reshape(masks.shape[0], masks.shape[1], 1)), axis=2)\n"
-			+ "        elif intersec == masks[:, :, overlapping[0]].sum() or intersec == masks[:, :, overlapping[1]].sum():\n"
-			+ "            pair += [[overlapping[0], overlapping[1]]]\n"
+			+ "            pairs = calculate_pairs(masks)\n"
+			+ "        elif intersec == masks[:, :, pp[0]].sum() or intersec == masks[:, :, pp[1]].sum():\n"
+			+ "            pairs = np.concatenate((pairs, np.unique(np.array([pp[0], pp[1]])).reshape(-1, 2)), axis=0)\n"
 			+ "        else:\n"
 			+ "            ## TODO run again precition\n"
-			+ "            if score[overlapping[0]] > score[overlapping[1]]:\n"
-			+ "                (masks[:, :, overlapping[1]] = ((masks[:, :, overlapping[1]] - masks[:, :, overlapping[0]]) > 0) * 1\n"
+			+ "            if score[pp[0]] > score[pp[1]]:\n"
+			+ "                (masks[:, :, pp[1]] = ((masks[:, :, pp[1]] - masks[:, :, pp[0]]) > 0) * 1\n"
+			+ "            else:\n"
+			+ "                (masks[:, :, pp[0]] = ((masks[:, :, pp[0]] - masks[:, :, pp[1]]) > 0) * 1\n"
+			+ "            pairs = calculate_pairs(masks)\n"
 			+ "        added_masks = masks.sum(2)\n"
 			+ "        inds = np.where(added_masks > 1)\n"
 			+ "    label = np.arange(masks.shape[2])\n"