Add example pipeline

Co-authored-by: Dolly Chen <[email protected]>
UCL · Nov 8, 2024 · c44fd94 · c44fd94
1 parent f4b2938
commit c44fd94
Show file tree

Hide file tree

Showing 5 changed files with 541 additions and 0 deletions.
diff --git a/example_pipeline/colon.yaml b/example_pipeline/colon.yaml
@@ -0,0 +1,12 @@
+
+
+# train and val data as 1) directory: path/images/, 2) file: path/images.txt, or 3) list: [path1/images/, path2/images/]
+train: /colon_data/polyps_small/images/train
+val: /colon_data/polyps_small/images/valid
+test: /colon_data/polyps_small/images/valid  # 20288 of 40670 images, submit to https://competitions.codalab.org/competitions/20794
+
+# number of classes
+nc: 1
+
+# class names
+names: [ 'polyp']
diff --git a/example_pipeline/find_bounding_box.ipynb b/example_pipeline/find_bounding_box.ipynb
@@ -0,0 +1,228 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Number of items: 2\n",
+      "Item 1: x_min=0, y_min=0, x_max=100, y_max=105\n",
+      "Item 2: x_min=144, y_min=225, x_max=193, y_max=278\n"
+     ]
+    }
+   ],
+   "source": [
+    "import cv2\n",
+    "import numpy as np\n",
+    "\n",
+    "def find_connected_components(image_path):\n",
+    "    # Read the image\n",
+    "    image = cv2.imread(image_path)\n",
+    "    \n",
+    "    # Convert the image to grayscale\n",
+    "    gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)\n",
+    "    \n",
+    "    # Threshold the image to create a binary image\n",
+    "    _, binary = cv2.threshold(gray, 1, 255, cv2.THRESH_BINARY)\n",
+    "    \n",
+    "    # Find connected components\n",
+    "    num_labels, labels, stats, centroids = cv2.connectedComponentsWithStats(binary, connectivity=8)\n",
+    "    \n",
+    "    # Initialize a list to store bounding boxes\n",
+    "    bounding_boxes = []\n",
+    "    \n",
+    "    for i in range(1, num_labels):  # Skip the background label 0\n",
+    "        x_min = stats[i, cv2.CC_STAT_LEFT]\n",
+    "        y_min = stats[i, cv2.CC_STAT_TOP]\n",
+    "        width = stats[i, cv2.CC_STAT_WIDTH]\n",
+    "        height = stats[i, cv2.CC_STAT_HEIGHT]\n",
+    "        x_max = x_min + width - 1\n",
+    "        y_max = y_min + height - 1\n",
+    "        \n",
+    "        bounding_boxes.append((x_min, y_min, x_max, y_max))\n",
+    "    \n",
+    "    return num_labels - 1, bounding_boxes  # Subtract 1 to exclude the background\n",
+    "\n",
+    "# Example usage\n",
+    "image_path = '../images/tst_seg_tst_seg500.png'\n",
+    "num_items, bounding_boxes = find_connected_components(image_path)\n",
+    "print(f\"Number of items: {num_items}\")\n",
+    "for i, (x_min, y_min, x_max, y_max) in enumerate(bounding_boxes):\n",
+    "    print(f\"Item {i+1}: x_min={x_min}, y_min={y_min}, x_max={x_max}, y_max={y_max}\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "metadata": {},
+   "outputs": [
+    {
+     "ename": "SyntaxError",
+     "evalue": "invalid syntax (2760322440.py, line 44)",
+     "output_type": "error",
+     "traceback": [
+      "\u001b[1;36m  Cell \u001b[1;32mIn[3], line 44\u001b[1;36m\u001b[0m\n\u001b[1;33m    num_items, boundinimport cv2\u001b[0m\n\u001b[1;37m                             ^\u001b[0m\n\u001b[1;31mSyntaxError\u001b[0m\u001b[1;31m:\u001b[0m invalid syntax\n"
+     ]
+    }
+   ],
+   "source": [
+    "import cv2\n",
+    "import numpy as np\n",
+    "\n",
+    "def find_connected_components(image_path, output_file):\n",
+    "    # Read the image\n",
+    "    image = cv2.imread(image_path)\n",
+    "    \n",
+    "    # Convert the image to grayscale\n",
+    "    gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)\n",
+    "    \n",
+    "    # Threshold the image to create a binary image\n",
+    "    _, binary = cv2.threshold(gray, 1, 255, cv2.THRESH_BINARY)\n",
+    "    \n",
+    "    # Find connected components\n",
+    "    num_labels, labels, stats, centroids = cv2.connectedComponentsWithStats(binary, connectivity=8)\n",
+    "    \n",
+    "    # Initialize a list to store bounding boxes\n",
+    "    bounding_boxes = []\n",
+    "    \n",
+    "    for i in range(1, num_labels):  # Skip the background label 0\n",
+    "        x_min = stats[i, cv2.CC_STAT_LEFT]\n",
+    "        y_min = stats[i, cv2.CC_STAT_TOP]\n",
+    "        width = stats[i, cv2.CC_STAT_WIDTH]\n",
+    "        height = stats[i, cv2.CC_STAT_HEIGHT]\n",
+    "        x_max = x_min + width - 1\n",
+    "        y_max = y_min + height - 1\n",
+    "        \n",
+    "        # Convert to center coordinates and dimensions\n",
+    "        x_center = x_min + width / 2\n",
+    "        y_center = y_min + height / 2\n",
+    "        \n",
+    "        bounding_boxes.append((x_center, y_center, width, height))\n",
+    "    \n",
+    "    # Write results to a text file\n",
+    "    with open(output_file, 'w') as f:\n",
+    "        for x_center, y_center, width, height in bounding_boxes:\n",
+    "            f.write(f\"0 {x_center:.8f} {y_center:.8f} {width:.8f} {height:.8f}\\n\")\n",
+    "    \n",
+    "    return num_labels - 1, bounding_boxes  # Subtract 1 to exclude the background\n",
+    "\n",
+    "# Example usage\n",
+    "image_path = '../images/tst_seg_tst_seg500.png'\n",
+    "output_file = '../labels/bounding_boxes.txt'\n",
+    "num_items, bounding_boxes = find_connected_components(image_path, output_file)\n",
+    "print(f\"Number of items: {num_items}\")\n",
+    "for i, (x_center, y_center, width, height) in enumerate(bounding_boxes):\n",
+    "    print(f\"Item {i+1}: x_center={x_center}, y_center={y_center}, width={width}, height={height}\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 4,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Number of items: 2\n",
+      "Item 1: x_center=0.09115523465703972, y_center=0.11830357142857142, width=0.18231046931407943, height=0.23660714285714285\n",
+      "Item 2: x_center=0.30505415162454874, y_center=0.5625, width=0.09025270758122744, height=0.12053571428571429\n"
+     ]
+    }
+   ],
+   "source": [
+    "import cv2\n",
+    "import numpy as np\n",
+    "import os\n",
+    "\n",
+    "def find_connected_components(image_path, output_dir):\n",
+    "    # Read the image\n",
+    "    image = cv2.imread(image_path)\n",
+    "    \n",
+    "    # Get image dimensions\n",
+    "    image_height, image_width = image.shape[:2]\n",
+    "    \n",
+    "    # Convert the image to grayscale\n",
+    "    gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)\n",
+    "    \n",
+    "    # Threshold the image to create a binary image\n",
+    "    _, binary = cv2.threshold(gray, 1, 255, cv2.THRESH_BINARY)\n",
+    "    \n",
+    "    # Find connected components\n",
+    "    num_labels, labels, stats, centroids = cv2.connectedComponentsWithStats(binary, connectivity=8)\n",
+    "    \n",
+    "    # Initialize a list to store bounding boxes\n",
+    "    bounding_boxes = []\n",
+    "    \n",
+    "    for i in range(1, num_labels):  # Skip the background label 0\n",
+    "        x_min = stats[i, cv2.CC_STAT_LEFT]\n",
+    "        y_min = stats[i, cv2.CC_STAT_TOP]\n",
+    "        width = stats[i, cv2.CC_STAT_WIDTH]\n",
+    "        height = stats[i, cv2.CC_STAT_HEIGHT]\n",
+    "        \n",
+    "        # Convert to center coordinates and dimensions\n",
+    "        x_center = x_min + width / 2\n",
+    "        y_center = y_min + height / 2\n",
+    "        \n",
+    "        # Convert to ratios\n",
+    "        x_center /= image_width\n",
+    "        y_center /= image_height\n",
+    "        width /= image_width\n",
+    "        height /= image_height\n",
+    "        \n",
+    "        bounding_boxes.append((x_center, y_center, width, height))\n",
+    "    \n",
+    "    # Create the output file path\n",
+    "    base_name = os.path.basename(image_path)\n",
+    "    output_file = os.path.join(output_dir, os.path.splitext(base_name)[0] + '.txt')\n",
+    "    \n",
+    "    # Write results to a text file\n",
+    "    with open(output_file, 'w') as f:\n",
+    "        for x_center, y_center, width, height in bounding_boxes:\n",
+    "            f.write(f\"0 {x_center:.8f} {y_center:.8f} {width:.8f} {height:.8f}\\n\")\n",
+    "    \n",
+    "    return num_labels - 1, bounding_boxes  # Subtract 1 to exclude the background\n",
+    "\n",
+    "# Example usage\n",
+    "image_path = '../images/tst_seg_tst_seg500.png'\n",
+    "output_dir = '../labels'\n",
+    "num_items, bounding_boxes = find_connected_components(image_path, output_dir)\n",
+    "print(f\"Number of items: {num_items}\")\n",
+    "for i, (x_center, y_center, width, height) in enumerate(bounding_boxes):\n",
+    "    print(f\"Item {i+1}: x_center={x_center}, y_center={y_center}, width={width}, height={height}\")"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "pgta",
+   "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": "3.11.10"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}
diff --git a/example_pipeline/instructions.md b/example_pipeline/instructions.md
@@ -0,0 +1,62 @@
+This is an example pipeline on how to convert output from Blender Randomiser to COCO format labelling in preperation of fine tuning polyps detection using YOLOv7
+
+# Data Preperation
+Once the segmentation of polyps are obtained. Use find_bounding_box.ipynb to output the bounding box information as text files.
+
+show_bbox.ipynb can be used to show if the correct bounding boxes have been converted.
+
+# IMPORTANT!
+**<span style="color:red">
+When Google Colab session finishes, ALL data is wiped, please save the models!
+</span>**
+## To tune the model using custom data:
+- clone yolov7 git and install it https://github.com/WongKinYiu/yolov7; pip install -r requirements.txt
+- log into wandb account (your own API - see below) DO NOT SHARE!
+- run python training script
+
+```python train.py --epochs 100 --device 0 --entity colon_coders --workers 8 --batch-size 32 --data /content/colon.yaml --img 512 512 --cfg /content/yolov7_training_config.yaml --weights '/content/yolov7_training.pt' --name yolov7-colon --hyp data/hyp.scratch.custom.yaml```
+
+## Google colab instructions
+- upload polyps.zip to google drive
+- upload colon.yaml, yolov7_training.pt, yolov7_training_config.yaml to google drive
+- open google colab and mount drive
+- unzip polyps.zip
+```python
+import zipfile
+with zipfile.ZipFile("/content/drive/MyDrive/polyps.zip", 'r') as zip_ref:
+    zip_ref.extractall("/content/colon_data")
+```
+- Important: remove the cache files (otherwise the model will use the cache file to load the data which has the incorrect file paths)
+- could use to code in show_bbox.ipynb to see if data and bounding boxes has been loaded correctly
+- install yolo7
+```python
+!git clone https://github.com/WongKinYiu/yolov7
+%cd yolov7
+!pip install -r requirements.txt
+```
+- set up wandb
+```python
+!pip install wandb
+import wandb
+wandb.login()
+```
+- tune model: make sure colon.yaml has the correct file paths for data, also make sure --data, --cfg and --weights has the correct file paths
+
+```python train.py --epochs 100 --device 0 --entity colon_coders --workers 8 --batch-size 32 --data /content/colon.yaml --img 512 512 --cfg /content/yolov7_training_config.yaml --weights '/content/yolov7_training.pt' --name yolov7-colon --hyp data/hyp.scratch.custom.yaml```
+- When training is finished, model output is saved under yolov7/runs/train
+
+## Run on test data
+!python test.py --data /content/colon.yaml --img 512 --batch 32 --conf 0.001 --iou 0.65 --device 0 --weights runs/train/yolov7-colon2/weights/best.pt --name yolov7_colon_val
+
+## Notes
+- The data location is specified in the config file colon.yaml
+- Training config is specified in yolov7_training_config.yaml
+
+## Weights and Biases
+Weights and Biases is a very good tool to use to track training progress. YoloV7 uses this tool and it is very easy to set up
+- https://wandb.ai/site/research sign up for the free account
+- log into your account, go to top right, under your name select "user profile"
+- go to section "danger zone" and reveal your API code
+- this code is then used to log in to wandb when prompted
+- when you finish you can change the API or throw it away.
+- when training is in progress, go to WandB website, click on top left, you should see the project YOLO which will show the current training session
diff --git a/example_pipeline/show_bbox.ipynb b/example_pipeline/show_bbox.ipynb