diff --git a/notebooks/saving_and_loading.ipynb b/notebooks/saving_and_loading.ipynb
index fd912d6..1b903db 100644
--- a/notebooks/saving_and_loading.ipynb
+++ b/notebooks/saving_and_loading.ipynb
@@ -2,7 +2,7 @@
  "cells": [
   {
    "cell_type": "code",
-   "execution_count": 1,
+   "execution_count": 5,
    "id": "f9a8c75d-11d7-4f4d-923a-054f10a60105",
    "metadata": {},
    "outputs": [],
@@ -13,10 +13,19 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 2,
+   "execution_count": 9,
    "id": "52306290-e9e9-4e96-9fe6-84a88e84b88d",
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "The autoreload extension is already loaded. To reload it, use:\n",
+      "  %reload_ext autoreload\n"
+     ]
+    }
+   ],
    "source": [
     "%load_ext autoreload\n",
     "%autoreload 3 -p\n",
@@ -42,7 +51,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 3,
+   "execution_count": 10,
    "id": "46ddbbf0-abfc-409f-b9c4-573232839472",
    "metadata": {},
    "outputs": [],
@@ -52,27 +61,18 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 4,
+   "execution_count": 15,
    "id": "008c7597-2e64-4ac9-b11a-938d33c3102c",
    "metadata": {},
-   "outputs": [
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "/home/ajkdrag/workspace/ocrtoolkit/src/ocrtoolkit/utilities/io_utils.py:7: TqdmExperimentalWarning: Using `tqdm.autonotebook.tqdm` in notebook mode. Use `tqdm.tqdm` instead to force console mode (e.g. in jupyter console)\n",
-      "  from tqdm.autonotebook import tqdm\n"
-     ]
-    }
-   ],
+   "outputs": [],
    "source": [
-    "from ocrtoolkit.core import detect\n",
+    "from ocrtoolkit.core import detect, detect_and_save_h5\n",
     "from ocrtoolkit.datasets import FileDS"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 9,
+   "execution_count": 12,
    "id": "0a561a4e-8b57-4d7d-822f-1baff89e5208",
    "metadata": {},
    "outputs": [],
@@ -90,7 +90,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 5,
+   "execution_count": 13,
    "id": "3f4c3a41-a659-409c-9c5b-199469e9030a",
    "metadata": {},
    "outputs": [
@@ -98,7 +98,7 @@
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "\u001b[32m2024-03-18 11:35:12.315\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mocrtoolkit.wrappers.model\u001b[0m:\u001b[36m__init__\u001b[0m:\u001b[36m22\u001b[0m - \u001b[1mLoaded model from ../pretrained/best.pt, to cpu\u001b[0m\n"
+      "\u001b[32m2024-03-30 20:51:41.466\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mocrtoolkit.wrappers.model\u001b[0m:\u001b[36m__init__\u001b[0m:\u001b[36m22\u001b[0m - \u001b[1mLoaded model from ../pretrained/best.pt, to cpu\u001b[0m\n"
      ]
     }
    ],
@@ -116,14 +116,14 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 6,
+   "execution_count": 14,
    "id": "9c1677f0-c574-43dc-af60-37a1e7684598",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "3c1b9667baa6472e9417093e9a2c1132",
+       "model_id": "383ccd8a0e964e1a8620d15b9a57235b",
        "version_major": 2,
        "version_minor": 0
       },
@@ -138,7 +138,7 @@
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "\u001b[32m2024-03-18 11:35:20.645\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mocrtoolkit.utilities.io_utils\u001b[0m:\u001b[36mget_files\u001b[0m:\u001b[36m57\u001b[0m - \u001b[1mFound 235 files.\u001b[0m\n"
+      "\u001b[32m2024-03-30 20:51:41.729\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mocrtoolkit.utilities.io_utils\u001b[0m:\u001b[36mget_files\u001b[0m:\u001b[36m57\u001b[0m - \u001b[1mFound 235 files.\u001b[0m\n"
      ]
     }
    ],
@@ -146,6 +146,38 @@
     "mini_ds = FileDS(\"../data/public/images/\").sample()"
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": 16,
+   "id": "9f41c860-517a-40f0-b76c-5a720661b56b",
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "8073008f3e33428488bf711937963187",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "  0%|          | 0/237 [00:00<?, ?it/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "\u001b[32m2024-03-30 20:53:03.613\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mocrtoolkit.utilities.io_utils\u001b[0m:\u001b[36mget_files\u001b[0m:\u001b[36m57\u001b[0m - \u001b[1mFound 235 files.\u001b[0m\n"
+     ]
+    }
+   ],
+   "source": [
+    "medium_ds = FileDS(\"../data/public/images/\").sample(k=50)"
+   ]
+  },
   {
    "cell_type": "markdown",
    "id": "47d1abff-d792-4104-b374-d4cbf71e4a62",
@@ -164,6 +196,56 @@
     "det_results = detect(detection_model, mini_ds, stream=False, verbose=False)"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "id": "a44b8e59-2c00-4c22-8e9a-7083c16291b0",
+   "metadata": {},
+   "source": [
+    "### saving detection results during inference"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 19,
+   "id": "86c13d81-5430-4777-96d8-f90266745bd0",
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "34f06fd9166b42f883c8cfaf7f79fd93",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "  0%|          | 0/10 [00:00<?, ?it/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "\u001b[32m2024-03-30 20:53:59.691\u001b[0m | \u001b[1mINFO    \u001b[0m | \u001b[36mocrtoolkit.utilities.det_utils\u001b[0m:\u001b[36msave_dets\u001b[0m:\u001b[36m17\u001b[0m - \u001b[1mDetections saved to temp/medium_dets_save.h5\u001b[0m\n"
+     ]
+    }
+   ],
+   "source": [
+    "detect_and_save_h5(detection_model, medium_ds, \"temp/medium_dets_save.h5\", bs=5, verbose=False)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 20,
+   "id": "d3afe405-7928-40c6-a038-7df75fb22fc9",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "loaded_medium_dets = load_dets(\"temp/medium_dets_save.h5\")"
+   ]
+  },
   {
    "cell_type": "markdown",
    "id": "2716e2d9-defa-4ab1-9bbd-722112be6b00",
diff --git a/src/ocrtoolkit/core/detector.py b/src/ocrtoolkit/core/detector.py
index 3e173b8..ae18f99 100644
--- a/src/ocrtoolkit/core/detector.py
+++ b/src/ocrtoolkit/core/detector.py
@@ -1,7 +1,9 @@
 import numpy as np
 from loguru import logger
+from tqdm.auto import tqdm
 
 from ocrtoolkit.datasets.base import BaseDS
+from ocrtoolkit.utilities.det_utils import save_dets
 from ocrtoolkit.wrappers.model import DetectionModel
 
 
@@ -34,3 +36,26 @@ def detect(model: DetectionModel, ds: BaseDS, stream=True, **kwargs):
     if stream:
         return gen
     return list(_detect(model, ds, **kwargs))
+
+
+def detect_and_save_h5(
+    model: DetectionModel,
+    ds: BaseDS,
+    path: str,
+    bs=4,
+    start_batch_idx=0,
+    **kwargs,
+):
+    """Detects objects in a dataset
+    Call model.preprocess methods before model.predict methods
+    Images should be converted to np.ndarray before calling preprocess
+    Saves the detections to path.
+    Caution: This captures all detection results in mem before saving.
+    """
+    num_batches = ds.num_batches(bs)
+    l_det_results = []
+    for idx in tqdm(range(start_batch_idx, num_batches)):
+        batch = ds.batch(bs, idx)
+        l_det_results += detect(model, batch, stream=False, **kwargs)
+
+    save_dets(l_det_results, path)