Upload 20 files

.gitattributes

@@ -33,3 +33,11 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+results/yolov8n-fashionpedia-1.torchscript filter=lfs diff=lfs merge=lfs -text
+results/yolov8s-fashionpedia-1.torchscript filter=lfs diff=lfs merge=lfs -text
+shared/yolov8n-fashionpedia-1.torchscript filter=lfs diff=lfs merge=lfs -text
+test/courte.png filter=lfs diff=lfs merge=lfs -text
+test/longue.png filter=lfs diff=lfs merge=lfs -text
+test/rayure.png filter=lfs diff=lfs merge=lfs -text
+test/sans-manche.png filter=lfs diff=lfs merge=lfs -text
+test/trois-quart.png filter=lfs diff=lfs merge=lfs -text

dataset-export-yolo.ipynb

@@ -0,0 +1,229 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "id": "9ca8370e29b99d9b",
+   "metadata": {
+    "collapsed": true,
+    "ExecuteTime": {
+     "end_time": "2024-04-17T07:40:35.575275Z",
+     "start_time": "2024-04-17T07:40:35.397474Z"
+    }
+   },
+   "source": [
+    "import json\n",
+    "import os\n",
+    "import shutil\n",
+    "\n",
+    "import pandas as pd"
+   ],
+   "execution_count": 2,
+   "outputs": []
+  },
+  {
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-04-17T07:59:30.461558Z",
+     "start_time": "2024-04-17T07:59:30.458532Z"
+    }
+   },
+   "cell_type": "code",
+   "source": [
+    "def load_dataset(input_annotations_file: str, input_img_dir: str, output_dir: str):\n",
+    "    with open(input_annotations_file, \"r\") as f:\n",
+    "        dataset = json.load(f)\n",
+    "    df_annotation = pd.DataFrame(dataset['annotations']).drop(['segmentation', 'attribute_ids'], axis='columns')\n",
+    "    df_images = pd.DataFrame(dataset['images'])\n",
+    "    df_categories = pd.DataFrame(dataset['categories'])\n",
+    "\n",
+    "    df_dataset = pd.merge(df_images, df_annotation, how='inner', left_on='id', right_on='image_id')\n",
+    "    df_dataset = pd.merge(df_dataset, df_categories, how='inner', left_on='category_id', right_on='id')\n",
+    "    df_dataset['path_to_image'] = input_img_dir + df_dataset['file_name']\n",
+    "    return df_dataset\n"
+   ],
+   "id": "initial_id",
+   "execution_count": 48,
+   "outputs": []
+  },
+  {
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-04-17T07:59:30.763506Z",
+     "start_time": "2024-04-17T07:59:30.759581Z"
+    }
+   },
+   "cell_type": "code",
+   "source": [
+    "def save_labels_one_image(df_one_image: pd.DataFrame, output_folder: str, subfolder: str):\n",
+    "    path_to_image = df_one_image.iloc[0]['path_to_image']\n",
+    "    image_id = df_one_image.iloc[0]['image_id']\n",
+    "    image_width = df_one_image.iloc[0]['width']\n",
+    "    image_height = df_one_image.iloc[0]['height']\n",
+    "\n",
+    "    def format_one_box(one_line) -> str:\n",
+    "        label_index = str(one_line['category_id'])\n",
+    "\n",
+    "        [x_center, y_center, width, height] = one_line['bbox']\n",
+    "        x_center_normalised, y_center_normalised = (x_center+width/2) / image_width, (y_center+height/2) / image_height\n",
+    "        width_normalised, height_normalised = width / image_width, height / image_height\n",
+    "        box = ' '.join(map(str, [x_center_normalised, y_center_normalised, width_normalised, height_normalised]))\n",
+    "\n",
+    "        line = f'{label_index} {box}'\n",
+    "        return line\n",
+    "\n",
+    "    boxes_formatted = df_one_image.apply(format_one_box, axis='columns').tolist()\n",
+    "\n",
+    "    with open(os.path.join(output_folder, 'labels', subfolder, f'{image_id}.txt'), 'w') as f:\n",
+    "        f.write('\\n'.join(boxes_formatted))\n",
+    "    shutil.copy(path_to_image, os.path.join(output_folder, 'images', subfolder, f'{image_id}.jpg'))\n",
+    "\n",
+    "    return image_id\n"
+   ],
+   "id": "aa0ec50c0b8b00f3",
+   "execution_count": 49,
+   "outputs": []
+  },
+  {
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-04-17T07:59:33.282829Z",
+     "start_time": "2024-04-17T07:59:32.281139Z"
+    }
+   },
+   "cell_type": "code",
+   "source": [
+    "path_to_json_validation = \"./dataset/fashionpedia2020/instances_attributes_val2020.json\"\n",
+    "path_to_image_folder_validation = './dataset/fashionpedia2020/val_test2020/test/'\n",
+    "output_folder_validation = 'val2020'\n",
+    "fashionpedia_dataset_validation = load_dataset(path_to_json_validation, path_to_image_folder_validation,\n",
+    "                                               'dataset/fashionpedia-yolo')\n",
+    "fashionpedia_dataset_validation.groupby(['image_id']).apply(save_labels_one_image, './dataset/fashionpedia-yolo',\n",
+    "                                                            output_folder_validation)"
+   ],
+   "id": "31756215355a324e",
+   "execution_count": 50,
+   "outputs": []
+  },
+  {
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-04-17T08:00:11.718962Z",
+     "start_time": "2024-04-17T07:59:33.326724Z"
+    }
+   },
+   "cell_type": "code",
+   "source": [
+    "path_to_json_train = \"./dataset/fashionpedia2020/instances_attributes_train2020.json\"\n",
+    "path_to_image_folder_train = './dataset/fashionpedia2020/train2020/train/'\n",
+    "output_folder_train = 'train2020'\n",
+    "fashionpedia_dataset_train = load_dataset(path_to_json_train, path_to_image_folder_train, 'dataset/fashionpedia-yolo')\n",
+    "fashionpedia_dataset_train.groupby(['image_id']).apply(save_labels_one_image, './dataset/fashionpedia-yolo',\n",
+    "                                                       output_folder_train)"
+   ],
+   "id": "ba95b509750ea10b",
+   "execution_count": 51,
+   "outputs": []
+  },
+  {
+   "metadata": {},
+   "cell_type": "code",
+   "execution_count": null,
+   "source": "",
+   "id": "77dff2cb00f9b08c",
+   "outputs": []
+  },
+  {
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-04-17T08:00:11.733525Z",
+     "start_time": "2024-04-17T08:00:11.731305Z"
+    }
+   },
+   "cell_type": "code",
+   "source": [
+    "from matplotlib import pyplot as plt\n",
+    "\n",
+    "COLORS = ['lightcoral', 'yellowgreen', 'darkturquoise', 'hotpink', 'mediumslateblue']\n",
+    "\n",
+    "\n",
+    "def plot_results(pil_img, boxes, class_labels):\n",
+    "    plt.figure(figsize=(16, 10))\n",
+    "    plt.imshow(pil_img)\n",
+    "    ax = plt.gca()\n",
+    "    colors = COLORS * 100\n",
+    "    for [xcenter, ycenter, xlength, ylength], c, label in zip(boxes, colors, class_labels):\n",
+    "        xmin = (xcenter-(xlength/2)) * pil_img.width\n",
+    "        xmax = (xcenter+(xlength/2)) * pil_img.width\n",
+    "        ymin = (ycenter-(ylength/2)) * pil_img.height\n",
+    "        ymax = (ycenter+(ylength/2)) * pil_img.height\n",
+    "        \n",
+    "        print(xmin, ymin)\n",
+    "        ax.add_patch(plt.Rectangle((xmin, ymin), xmax - xmin, ymax - ymin,\n",
+    "                                   fill=False, color=c, linewidth=2))\n",
+    "        ax.text(xmin, ymin, str(label), fontsize=15,\n",
+    "                bbox=dict(facecolor=c, alpha=0.5))\n",
+    "    plt.axis('off')\n",
+    "    plt.show()"
+   ],
+   "id": "a16ebfa1cf517c1a",
+   "execution_count": 52,
+   "outputs": []
+  },
+  {
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-04-17T08:00:11.921853Z",
+     "start_time": "2024-04-17T08:00:11.733970Z"
+    }
+   },
+   "cell_type": "code",
+   "source": [
+    "from PIL import Image\n",
+    "\n",
+    "pil_image = Image.open('dataset/fashionpedia-yolo/images/train2020/23.jpg')\n",
+    "with open('dataset/fashionpedia-yolo/labels/train2020/23.txt' ) as f:\n",
+    "    labels = f.readlines()\n",
+    "\n",
+    "class_labels  = []\n",
+    "box_center_length = []\n",
+    "for a_label in labels:\n",
+    "    labels_split = a_label.split(' ')\n",
+    "    class_labels.append(int(labels_split[0]))\n",
+    "    box_center_length.append(list(map(float, labels_split[1:])))\n",
+    "plot_results(pil_image,box_center_length, class_labels)"
+   ],
+   "id": "8c5dacbcecca7d42",
+   "execution_count": 53,
+   "outputs": []
+  },
+  {
+   "metadata": {},
+   "cell_type": "code",
+   "execution_count": null,
+   "source": "",
+   "id": "d7ac9abea09107e1",
+   "outputs": []
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 2
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython2",
+   "version": "2.7.6"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}

dataset-stats.ipynb

@@ -0,0 +1,201 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "id": "755f1267a6176c24",
+   "metadata": {
+    "collapsed": true,
+    "ExecuteTime": {
+     "end_time": "2024-04-15T08:56:52.063622Z",
+     "start_time": "2024-04-15T08:56:51.856029Z"
+    }
+   },
+   "source": [
+    "import json\n",
+    "import pandas as pd"
+   ],
+   "execution_count": 1,
+   "outputs": []
+  },
+  {
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-04-15T08:57:05.764152Z",
+     "start_time": "2024-04-15T08:57:05.390988Z"
+    }
+   },
+   "cell_type": "code",
+   "source": [
+    "with open(\"./dataset/instances_attributes_val2020.json\", \"r\") as f:\n",
+    "    dataset = json.load(f)\n",
+    "\n",
+    "print(dataset.keys())"
+   ],
+   "id": "initial_id",
+   "execution_count": 3,
+   "outputs": []
+  },
+  {
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-04-15T09:41:18.117520Z",
+     "start_time": "2024-04-15T09:41:18.110946Z"
+    }
+   },
+   "cell_type": "code",
+   "source": [
+    "a = dataset['images']\n",
+    "print(len(a))\n",
+    "print(a)"
+   ],
+   "id": "24c0ff8dda428023",
+   "execution_count": 30,
+   "outputs": []
+  },
+  {
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-04-15T12:48:49.207137Z",
+     "start_time": "2024-04-15T12:48:49.202422Z"
+    }
+   },
+   "cell_type": "code",
+   "source": [
+    "df_categories = pd.DataFrame(dataset['categories'])\n",
+    "df_categories"
+   ],
+   "id": "4e45dd93ed3d6d11",
+   "execution_count": 107,
+   "outputs": []
+  },
+  {
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-04-15T09:02:58.490674Z",
+     "start_time": "2024-04-15T09:02:58.485590Z"
+    }
+   },
+   "cell_type": "code",
+   "source": [
+    "df_attributes = pd.DataFrame(dataset['attributes'])\n",
+    "df_attributes"
+   ],
+   "id": "f8c1f1649b4890b7",
+   "execution_count": 18,
+   "outputs": []
+  },
+  {
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-04-15T09:02:21.405195Z",
+     "start_time": "2024-04-15T09:02:21.390532Z"
+    }
+   },
+   "cell_type": "code",
+   "source": [
+    "df_annotation = pd.DataFrame(dataset['annotations'])\n",
+    "df_annotation = df_annotation.drop(['segmentation'], axis='columns')\n",
+    "df_annotation_exploded =df_annotation.explode('attribute_ids') "
+   ],
+   "id": "4c53c3115c3dbef8",
+   "execution_count": 16,
+   "outputs": []
+  },
+  {
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-04-15T09:02:23.057097Z",
+     "start_time": "2024-04-15T09:02:23.051760Z"
+    }
+   },
+   "cell_type": "code",
+   "source": "df_annotation_exploded",
+   "id": "618f069b857225fc",
+   "execution_count": 17,
+   "outputs": []
+  },
+  {
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-04-15T09:03:53.975762Z",
+     "start_time": "2024-04-15T09:03:53.959100Z"
+    }
+   },
+   "cell_type": "code",
+   "source": [
+    "df_joined = df_attributes.merge(df_annotation_exploded, how='left', left_on='id', right_on='attribute_ids')\n",
+    "df_joined"
+   ],
+   "id": "8a85bbf68e23920b",
+   "execution_count": 20,
+   "outputs": []
+  },
+  {
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-04-15T09:08:05.647378Z",
+     "start_time": "2024-04-15T09:08:05.637874Z"
+    }
+   },
+   "cell_type": "code",
+   "source": [
+    "df_neck_collar = df_joined[df_joined.name.str.contains('neck') | df_joined.name.str.contains('collar') ]\n",
+    "df_neck_collar['name'].value_counts()"
+   ],
+   "id": "2888345540b560d5",
+   "execution_count": 28,
+   "outputs": []
+  },
+  {
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-04-15T09:49:34.287590Z",
+     "start_time": "2024-04-15T09:49:34.263905Z"
+    }
+   },
+   "cell_type": "code",
+   "source": [
+    "df_annotation = pd.DataFrame(dataset['annotations']).drop(['segmentation', 'attribute_ids'], axis='columns')\n",
+    "df_images = pd.DataFrame(dataset['images'])\n",
+    "df_categories = pd.DataFrame(dataset['categories'])\n",
+    "\n",
+    "df_dataset = pd.merge(df_images, df_annotation, how='inner', left_on='id', right_on='image_id')\n",
+    "df_dataset = pd.merge(df_dataset, df_categories, how='inner', left_on='category_id', right_on='id')\n",
+    "df_dataset['path_to_image'] = 'coucou/' + df_dataset['file_name']\n",
+    "df_dataset"
+   ],
+   "id": "ebb14b7642a1459d",
+   "execution_count": 37,
+   "outputs": []
+  },
+  {
+   "metadata": {},
+   "cell_type": "code",
+   "execution_count": null,
+   "source": "",
+   "id": "20ba29a56c257d56",
+   "outputs": []
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 2
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython2",
+   "version": "2.7.6"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}

export_prediction_to_label_studio.py

@@ -0,0 +1,102 @@
+import json
+import os
+from uuid import uuid4
+
+
+def export_one_prediction_to_label_studio(prediction, label_studio_relative_path: str, output_folder: str,
+                                          model_name: str, from_name: str,
+                                          threshold_to_export: float = 0.1):
+    image_name = label_studio_relative_path.split('/')[-1]
+    output_file = image_name.replace('.jpg', ".json").replace('.jpeg', ".json")
+    output_path = os.path.join(output_folder, output_file)
+
+    original_width, original_height = prediction.orig_shape
+    boxes_filtered = filter(lambda x: float(x.conf[0]) > threshold_to_export, prediction.boxes)
+    json_content = {
+        "data": {
+            "img": label_studio_relative_path,
+        },
+        "predictions": [
+            {
+                "model_version": model_name,
+                "result": [
+                    {
+                        "original_width": original_width,
+                        "original_height": original_height,
+                        "score": float(box.conf[0]),
+                        "value": {
+                            # nightmareeeeeee
+                            "x": float(box.xywhn[0][0] - box.xywhn[0][2] / 2) * 100,
+                            "y": float(box.xywhn[0][1] - box.xywhn[0][3] / 2) * 100,
+                            "width": float(box.xywhn[0][2]) * 100,
+                            "height": float(box.xywhn[0][3]) * 100,
+                            "rotation": 0
+                        },
+                        "id": str(uuid4()),
+                        "from_name": from_name,
+                        "to_name": "image",
+                        "type": "rectangle"
+                    } for box in boxes_filtered],
+            }]
+    }
+
+    with open(output_path, 'w') as fp:
+        json.dump(json_content, fp)
+
+    return json_content
+
+
+def export_one_yolo_output_to_label_studio_rectangles(yolo_prediction,
+                                                      model_version: str,
+                                                      label_studio_relative_path: str,
+                                                      output_folder: str,
+                                                      label_studio_to_name: str = 'image',
+                                                      label_studio_from_name: str = 'label',
+                                                      confidence_threshold: float = 0.1,
+                                                      overwrite_existing_files: bool = False):
+    os.makedirs(output_folder, exist_ok=True)
+
+    image_name = label_studio_relative_path.split('/')[-1]
+    output_file = image_name.replace('.jpg', ".json").replace('.jpeg', ".json")
+    output_path = os.path.join(output_folder, output_file)
+
+    original_width, original_height = yolo_prediction.orig_shape
+    class_names = yolo_prediction.names
+    boxes_filtered = filter(lambda x: float(x.conf[0]) >= confidence_threshold
+                            , yolo_prediction.boxes)
+
+    if not os.path.exists(output_path) or overwrite_existing_files:
+        json_content = {
+            "data": {
+                "image": label_studio_relative_path,
+            },
+            "predictions": [
+                {
+                    "model_version": model_version,
+                    "result": [
+                        {
+                            "original_width": original_width,
+                            "original_height": original_height,
+                            "score": float(box.conf[0]),
+                            "value": {
+                                # label studio x, y, width and height are in percent multiplied by 100
+                                # x and y are the position of the box top left corner
+                                "x": float(box.xywhn[0][0] - box.xywhn[0][2] / 2) * 100,
+                                "y": float(box.xywhn[0][1] - box.xywhn[0][3] / 2) * 100,
+                                "width": float(box.xywhn[0][2]) * 100,
+                                "height": float(box.xywhn[0][3]) * 100,
+                                "rotation": 0,
+                                "rectanglelabels": [class_names[int(box.cls)]]
+                            },
+                            "id": str(uuid4()),
+                            "from_name": label_studio_from_name,
+                            "to_name": label_studio_to_name,
+                            "type": "rectanglelabels",
+                        } for box in boxes_filtered],
+                }]
+        }
+
+        with open(output_path, 'w') as fp:
+            json.dump(json_content, fp)
+
+        return json_content

fashionpedia.yaml

@@ -0,0 +1,54 @@
+path: fashionpedia-yolo
+train: images/train2020
+val: images/val2020
+test: # test images (optional)
+
+# Classes
+names:
+  0: "shirt, blouse"
+  1: "top, t-shirt, sweatshirt"
+  2: sweater
+  3: cardigan
+  4: jacket
+  5: vest
+  6: pants
+  7: shorts
+  8: skirt
+  9: coat
+  10: dress
+  11: jumpsuit
+  12: cape
+  13: glasses
+  14: hat
+  15: "headband, head covering, hair accessory"
+  16: tie
+  17: glove
+  18: watch
+  19: belt
+  20: leg warmer
+  21: "tights, stockings"
+  22: sock
+  23: shoe
+  24: "bag, wallet"
+  25: scarf
+  26: umbrella
+  27: hood
+  28: collar
+  29: lapel
+  30: epaulette
+  31: sleeve
+  32: pocket
+  33: neckline
+  34: buckle
+  35: zipper
+  36: applique
+  37: bead
+  38: bow
+  39: flower
+  40: fringe
+  41: ribbon
+  42: rivet
+  43: ruffle
+  44: sequin
+  45: tassel
+

results/yolov8n-fashionpedia-1.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6b581af3abe2b4ae97a84cbe9aef5ae37207a68ee966d22b0b8d7969d79e6920
+size 12273953

results/yolov8n-fashionpedia-1.torchscript

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2053a5d7f854e46a06394b6866b15427c3055bdbca95f94821a359c76e6a7b3e
+size 12505310

results/yolov8s-fashionpedia-1.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9432e4a850d0a3049724c897fc22bcedf5caa32092f7d023df04fa6c8823299e
+size 44789121

results/yolov8s-fashionpedia-1.torchscript

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:40667a8e28ae5b8d6bb30663de642467759536565fa57b959d78c7fcb760087e
+size 45020318

results/yolovn8-fashionpedia-1.md

@@ -0,0 +1,55 @@
+Epoch    GPU_mem   box_loss   cls_loss   dfl_loss  Instances       Size
+200/200       5.6G     0.7991     0.6805      1.012        181        640: 100%|██████████| 1426/1426 [02:24<00:00,  9.90it/s]
+Class     Images  Instances      Box(P          R      mAP50  mAP50-95): 100%|██████████| 713/713 [01:24<00:00,  8.46it/s]
+all      45623     333400      0.737      0.583      0.631      0.501
+200 epochs completed in 12.290 hours. 
+
+Class     Images  Instances      Box(P          R      mAP50  mAP50-95): 100%|██████████| 713/713 [01:28<00:00,  8.08it/s]
+all      45623     333400      0.735      0.583      0.631      0.501
+shirt, blouse      45623       6161      0.695      0.777      0.824      0.713
+top, t-shirt, sweatshirt      45623      16548      0.827      0.839      0.895      0.794
+sweater      45623       1494      0.749      0.775      0.828      0.772
+cardigan      45623       1107      0.761      0.836      0.868      0.807
+jacket      45623       7833      0.876      0.938      0.962      0.916
+vest      45623        719      0.683      0.565      0.648      0.477
+pants      45623      12414      0.868      0.961      0.974      0.921
+shorts      45623       2756      0.844      0.897      0.929      0.818
+skirt      45623       5046      0.793       0.91      0.933       0.87
+coat      45623       3124      0.815      0.916      0.945      0.876
+dress      45623      18739      0.921      0.965      0.985      0.941
+jumpsuit      45623        922      0.791      0.826      0.882      0.821
+cape      45623        152      0.763      0.914      0.909      0.838
+glasses      45623       4855      0.911      0.845      0.893      0.664
+hat      45623       2518      0.847      0.911      0.935      0.814
+headband, head covering, hair accessory      45623       3470        0.7      0.373      0.438      0.272
+tie      45623       1457      0.743      0.781      0.823      0.538
+glove      45623       1385      0.746      0.669       0.74      0.552
+watch      45623       3389      0.679      0.569      0.644      0.406
+belt      45623       6851      0.673      0.535      0.595      0.386
+leg warmer      45623        112      0.713      0.446       0.52      0.373
+tights, stockings      45623       4326      0.757      0.809      0.839      0.677
+sock      45623       2582      0.708      0.368      0.444      0.281
+shoe      45623      46374      0.899      0.909      0.946      0.757
+bag, wallet      45623       7217      0.755      0.751      0.801      0.595
+scarf      45623       1374      0.651      0.571      0.623      0.426
+umbrella      45623        135      0.744      0.607      0.687      0.556
+hood      45623       1226      0.662       0.62      0.649      0.486
+collar      45623      10159      0.817      0.751      0.826      0.603
+lapel      45623       5972      0.847      0.802      0.871      0.674
+epaulette      45623        874      0.644      0.399      0.468      0.264
+sleeve      45623      59448      0.909      0.907      0.946      0.779
+pocket      45623      27179      0.721      0.469      0.556      0.336
+neckline      45623      34258      0.804       0.75      0.814      0.554
+buckle      45623       3300       0.63      0.361      0.397      0.196
+zipper      45623       7991      0.609      0.154      0.235      0.104
+applique      45623       3529      0.515      0.154       0.21      0.133
+bead      45623       5084      0.529      0.145      0.159      0.105
+bow      45623        528       0.55      0.134      0.205      0.129
+flower      45623       1367      0.663      0.266      0.338      0.217
+fringe      45623        588      0.578      0.259      0.303      0.206
+ribbon      45623        274      0.395     0.0693       0.11     0.0747
+rivet      45623       4892          1     0.0015    0.00799     0.0041
+ruffle      45623       2407      0.567      0.295       0.36      0.266
+sequin      45623        929      0.464     0.0243     0.0753     0.0537
+tassel      45623        335          1          0    0.00828    0.00475
+Speed: 0.1ms preprocess, 0.4ms inference, 0.0ms loss, 0.3ms postprocess per image

results/yolovs8-fashionpedia-1.md

@@ -0,0 +1,55 @@
+Epoch GPU_mem box_loss cls_loss dfl_loss Instances Size
+200/200 7.76G 0.6451 0.4508 0.9281 181 640: 100%|██████████| 1426/1426 [03:52<00:00, 6.12it/s]
+Class Images Instances Box(P R mAP50 mAP50-95): 100%|██████████| 713/713 [01:43<00:00, 6.92it/s]
+all 45623 333400 0.852 0.737 0.79 0.661
+200 epochs completed in 18.949 hours
+
+Class     Images  Instances      Box(P          R      mAP50  mAP50-95): 100%|██████████| 713/713 [01:50<00:00,  6.43it/s]
+all      45623     333400      0.852      0.737       0.79      0.661
+shirt, blouse      45623       6161      0.857      0.887      0.941      0.863
+top, t-shirt, sweatshirt      45623      16548      0.939       0.92      0.964      0.898
+sweater      45623       1494      0.936      0.915      0.961      0.915
+cardigan      45623       1107      0.974      0.948      0.977      0.946
+jacket      45623       7833      0.982      0.976      0.991       0.97
+vest      45623        719      0.868      0.847      0.917      0.766
+pants      45623      12414       0.94      0.985      0.991      0.969
+shorts      45623       2756       0.95      0.965      0.984      0.915
+skirt      45623       5046      0.961       0.98      0.992      0.961
+coat      45623       3124      0.977       0.98      0.993      0.964
+dress      45623      18739      0.988      0.988      0.995      0.978
+jumpsuit      45623        922      0.972      0.971      0.991      0.962
+cape      45623        152      0.948      0.987      0.992      0.955
+glasses      45623       4855      0.946       0.91      0.945      0.749
+hat      45623       2518      0.929      0.963       0.98      0.905
+headband, head covering, hair accessory      45623       3470      0.819      0.533      0.616      0.441
+tie      45623       1457      0.848      0.898       0.93      0.692
+glove      45623       1385      0.899      0.853      0.918      0.747
+watch      45623       3389        0.8      0.709      0.803       0.55
+belt      45623       6851      0.826      0.721      0.806      0.591
+leg warmer      45623        112      0.894      0.848      0.876      0.723
+tights, stockings      45623       4326      0.927      0.921      0.958      0.838
+sock      45623       2582      0.869      0.542       0.64      0.448
+shoe      45623      46374      0.958      0.942      0.973      0.829
+bag, wallet      45623       7217      0.906      0.889      0.938      0.791
+scarf      45623       1374      0.854      0.811      0.887      0.715
+umbrella      45623        135      0.906      0.844      0.893      0.789
+hood      45623       1226      0.843       0.82      0.882      0.735
+collar      45623      10159      0.903      0.849      0.913      0.725
+lapel      45623       5972      0.914      0.888      0.942      0.788
+epaulette      45623        874       0.77      0.594      0.675       0.42
+sleeve      45623      59448      0.954      0.945      0.973      0.851
+pocket      45623      27179      0.837       0.63       0.74      0.491
+neckline      45623      34258      0.894      0.821       0.89      0.661
+buckle      45623       3300      0.764      0.466      0.545      0.307
+zipper      45623       7991      0.742      0.342      0.454       0.26
+applique      45623       3529      0.708      0.474      0.545      0.383
+bead      45623       5084      0.674      0.237      0.283      0.205
+bow      45623        528      0.737      0.505      0.611      0.448
+flower      45623       1367      0.778      0.538      0.628      0.441
+fringe      45623        588      0.722      0.602       0.65      0.497
+ribbon      45623        274      0.692       0.31      0.393      0.304
+rivet      45623       4892      0.564      0.018     0.0309     0.0108
+ruffle      45623       2407      0.792      0.604      0.715      0.583
+sequin      45623        929      0.712      0.333      0.404      0.311
+tassel      45623        335      0.528       0.18      0.228      0.131
+Speed

shared/describe_clothes.py

@@ -0,0 +1,32 @@
+import json
+import os
+from typing import List
+
+from numpy import ndarray
+from ultralytics.models import YOLO
+
+
+def import_describe_model() -> YOLO:
+    current_folder = os.path.realpath(os.path.join(os.getcwd(), os.path.dirname(__file__)))
+    neural_network_file = os.path.join(current_folder, "./yolov8n-fashionpedia-1.torchscript")
+    return YOLO(neural_network_file, task='detect')
+
+
+def describe_clothes_batch_opencv_bgr(yolo_model_loaded: YOLO,
+                                      pictures_rgb: List[ndarray],
+                                      threshold_to_save: float):
+    results = yolo_model_loaded(pictures_rgb, verbose=False, conf=threshold_to_save)
+
+    formatted_results = []
+    for a_result in results:
+        formatted_results.append(json.loads(a_result.tojson()))
+
+    return formatted_results
+
+
+def describe_single_clothes_opencv_rgb(yolo_model_loaded: YOLO,
+                                       one_clothes_picture_rgb: ndarray,
+                                       threshold_to_save: float):
+    return describe_clothes_batch_opencv_bgr(yolo_model_loaded,
+                                             [one_clothes_picture_rgb],
+                                             threshold_to_save)[0]

shared/yolov8n-fashionpedia-1.torchscript

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2053a5d7f854e46a06394b6866b15427c3055bdbca95f94821a359c76e6a7b3e
+size 12505310

test/test_shared.py

@@ -0,0 +1,17 @@
+import cv2
+
+from clothes_description_v1.shared.describe_clothes import import_describe_model, describe_clothes_batch_opencv_bgr, \
+    describe_single_clothes_opencv_bgr
+
+
+def test_shared_opencv():
+    yolo_model = import_describe_model()
+    courte = cv2.imread('courte.png')
+    results = describe_clothes_batch_opencv_bgr(yolo_model, [courte])
+    print(results)
+    result = describe_single_clothes_opencv_bgr(yolo_model, courte)
+    print(result)
+
+
+if __name__ == '__main__':
+    test_shared_opencv()

yolo8-pre-annotate.ipynb

@@ -0,0 +1,315 @@
+{
+ "cells": [
+  {
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-08-28T07:38:11.097948Z",
+     "start_time": "2024-08-28T07:38:10.051274Z"
+    }
+   },
+   "cell_type": "code",
+   "source": [
+    "import json\n",
+    "\n",
+    "from ultralytics import YOLO\n",
+    "import os\n",
+    "\n",
+    "from export_prediction_to_label_studio import export_one_yolo_output_to_label_studio_rectangles"
+   ],
+   "id": "688b209f5c7b833e",
+   "outputs": [],
+   "execution_count": 1
+  },
+  {
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-08-28T07:38:24.127255Z",
+     "start_time": "2024-08-28T07:38:23.889160Z"
+    }
+   },
+   "cell_type": "code",
+   "source": [
+    "%load_ext autoreload\n",
+    "%autoreload 2"
+   ],
+   "id": "bcbe8835eff23931",
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "The autoreload extension is already loaded. To reload it, use:\n",
+      "  %reload_ext autoreload\n"
+     ]
+    }
+   ],
+   "execution_count": 3
+  },
+  {
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-08-28T07:38:24.905478Z",
+     "start_time": "2024-08-28T07:38:24.890872Z"
+    }
+   },
+   "cell_type": "code",
+   "source": "yolo_model_loaded = YOLO(\"shared/yolov8n-fashionpedia-1.torchscript\", task='detect')",
+   "id": "f427e4ba2cff784",
+   "outputs": [],
+   "execution_count": 4
+  },
+  {
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-08-28T07:38:31.339742Z",
+     "start_time": "2024-08-28T07:38:31.325405Z"
+    }
+   },
+   "cell_type": "code",
+   "source": "root_dataset_folder = '/home/louis/vetement_ai/ai_that_look/pattern_detection/dataset/dataset-clothes-8-pattern-v0-labeled-15k'",
+   "id": "f6b614baa08e709",
+   "outputs": [],
+   "execution_count": 5
+  },
+  {
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-08-28T07:42:03.489627Z",
+     "start_time": "2024-08-28T07:39:47.352446Z"
+    }
+   },
+   "cell_type": "code",
+   "source": [
+    "for root, dirs, files in os.walk(os.path.join(root_dataset_folder, 'images')):\n",
+    "    print(root)\n",
+    "    for a_file in files:\n",
+    "        if a_file.endswith('.jpg') or a_file.endswith('.jpeg') or a_file.endswith('.png'):\n",
+    "            result = yolo_model_loaded([os.path.join(root, a_file)], verbose=False)[0]\n",
+    "            current_folder = root.split('/')[-1]\n",
+    "            export_one_yolo_output_to_label_studio_rectangles(\n",
+    "                result,\n",
+    "                'yolov8n-fashionpedia-1-torchscript-c98616b267958c24874073ce091a977ce25489f5',\n",
+    "                f'/data/local-files/?d=clothes_pattern/images/{current_folder}/{a_file}',\n",
+    "                root.replace('images/', 'pre-annotations/'),\n",
+    "                label_studio_from_name='clothes_pattern'\n",
+    "            )"
+   ],
+   "id": "66db1d64f1f26295",
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "/home/louis/vetement_ai/ai_that_look/pattern_detection/dataset/dataset-clothes-8-pattern-v0-labeled-15k/images\n",
+      "/home/louis/vetement_ai/ai_that_look/pattern_detection/dataset/dataset-clothes-8-pattern-v0-labeled-15k/images/plisse_drape_fronce\n",
+      "/home/louis/vetement_ai/ai_that_look/pattern_detection/dataset/dataset-clothes-8-pattern-v0-labeled-15k/images/jersey cotele\n",
+      "/home/louis/vetement_ai/ai_that_look/pattern_detection/dataset/dataset-clothes-8-pattern-v0-labeled-15k/images/non_classe\n",
+      "/home/louis/vetement_ai/ai_that_look/pattern_detection/dataset/dataset-clothes-8-pattern-v0-labeled-15k/images/rayure\n",
+      "/home/louis/vetement_ai/ai_that_look/pattern_detection/dataset/dataset-clothes-8-pattern-v0-labeled-15k/images/motif repete\n",
+      "/home/louis/vetement_ai/ai_that_look/pattern_detection/dataset/dataset-clothes-8-pattern-v0-labeled-15k/images/liberty\n",
+      "/home/louis/vetement_ai/ai_that_look/pattern_detection/dataset/dataset-clothes-8-pattern-v0-labeled-15k/images/motif place\n",
+      "/home/louis/vetement_ai/ai_that_look/pattern_detection/dataset/dataset-clothes-8-pattern-v0-labeled-15k/images/pied de poule\n",
+      "/home/louis/vetement_ai/ai_that_look/pattern_detection/dataset/dataset-clothes-8-pattern-v0-labeled-15k/images/animal\n",
+      "/home/louis/vetement_ai/ai_that_look/pattern_detection/dataset/dataset-clothes-8-pattern-v0-labeled-15k/images/pois\n",
+      "/home/louis/vetement_ai/ai_that_look/pattern_detection/dataset/dataset-clothes-8-pattern-v0-labeled-15k/images/pois_monochrome\n",
+      "/home/louis/vetement_ai/ai_that_look/pattern_detection/dataset/dataset-clothes-8-pattern-v0-labeled-15k/images/carreau\n",
+      "/home/louis/vetement_ai/ai_that_look/pattern_detection/dataset/dataset-clothes-8-pattern-v0-labeled-15k/images/paillette\n",
+      "/home/louis/vetement_ai/ai_that_look/pattern_detection/dataset/dataset-clothes-8-pattern-v0-labeled-15k/images/floral\n"
+     ]
+    }
+   ],
+   "execution_count": 8
+  },
+  {
+   "metadata": {},
+   "cell_type": "code",
+   "outputs": [],
+   "execution_count": null,
+   "source": "",
+   "id": "513093f0ca4ac231"
+  },
+  {
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-08-07T14:10:47.957914Z",
+     "start_time": "2024-08-07T14:10:47.920287Z"
+    }
+   },
+   "cell_type": "code",
+   "source": [
+    "kids_output = [\n",
+    "    \"body\",\n",
+    "    \"chemise\",\n",
+    "    \"combinaison\",\n",
+    "    \"gigoteuse\",\n",
+    "    \"gilet\",\n",
+    "    \"jupe\",\n",
+    "    \"pantalon\",\n",
+    "    \"pyjama\",\n",
+    "    \"robe\",\n",
+    "    \"tshirt\"\n",
+    "]\n",
+    "\n",
+    "attributes_to_transform_default = {\n",
+    "    \"shirt, blouse\": \"chemise\",\n",
+    "    \"top, t-shirt, sweatshirt\": \"tshirt\",\n",
+    "    \"sweater\": \"gilet\",\n",
+    "    \"cardigan\": \"gilet\",\n",
+    "    \"jacket\": \"gilet\",\n",
+    "    \"vest\": \"gilet\",\n",
+    "    \"pants\": \"pantalon\",\n",
+    "    \"shorts\": \"pantalon\",\n",
+    "    \"skirt\": \"jupe\",\n",
+    "    \"coat\": \"gilet\",\n",
+    "    \"dress\": \"robe\",\n",
+    "    \"jumpsuit\": \"combinaison\",\n",
+    "}\n",
+    "attributes_to_keep = {\n",
+    "    \"sleeve\": \"manche\",\n",
+    "    \"pocket\": \"poche\",\n",
+    "    \"buckle\": \"zip\",\n",
+    "    \"collar\": \"col\",\n",
+    "    \"lapel\": \"revers\",\n",
+    "    \"hood\": \"capuche\",\n",
+    "}"
+   ],
+   "id": "aa14e86f94358cb1",
+   "outputs": [],
+   "execution_count": 27
+  },
+  {
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-08-07T14:53:59.250707Z",
+     "start_time": "2024-08-07T14:53:57.334490Z"
+    }
+   },
+   "cell_type": "code",
+   "source": [
+    "attributes_to_transform = {k: 'pyjama' for k in attributes_to_transform_default.keys()}\n",
+    "\n",
+    "attributes_to_label = {\n",
+    "    **attributes_to_transform,\n",
+    "    **attributes_to_keep,\n",
+    "}\n",
+    "\n",
+    "assert len(set(attributes_to_transform.keys()).intersection(set(attributes_to_transform.values()))) == 0\n",
+    "\n",
+    "score_threshold = 0.5\n",
+    "for root, dirs, files in os.walk(os.path.join(root_dataset_folder, 'pre-annotations', 'pyjama')):\n",
+    "    for a_file in files:\n",
+    "        if a_file.endswith('.json'):\n",
+    "            filepath = os.path.join(root, a_file)\n",
+    "            with open(filepath, 'r') as f:\n",
+    "                file_content_json = json.load(f)\n",
+    "\n",
+    "            predictions = file_content_json['predictions']\n",
+    "            annotations = predictions[0]['result']\n",
+    "            annotations = [p for p in annotations if p['score'] >= score_threshold]\n",
+    "\n",
+    "            updated_annotations = []\n",
+    "            for an_annotation in annotations:\n",
+    "                rectangle_labels = an_annotation['value']['rectanglelabels']\n",
+    "\n",
+    "                updated_labels = []\n",
+    "                for a_label in rectangle_labels:\n",
+    "                    if a_label in attributes_to_label.keys():\n",
+    "                        updated_labels.append(attributes_to_label[a_label])\n",
+    "                    elif a_label in attributes_to_label.values():\n",
+    "                        # to make it idempotent\n",
+    "                        updated_labels.append(a_label)\n",
+    "\n",
+    "                if len(updated_labels) > 0:\n",
+    "                    an_annotation['value']['rectanglelabels'] = updated_labels\n",
+    "                    updated_annotations.append(an_annotation)\n",
+    "\n",
+    "            with open(filepath, 'w') as f:\n",
+    "                json.dump({\n",
+    "                    **file_content_json,\n",
+    "                    'predictions': [\n",
+    "                        {\n",
+    "                            **predictions[0],\n",
+    "                            \"result\": updated_annotations\n",
+    "                        }\n",
+    "                    ]\n",
+    "                }, f)"
+   ],
+   "id": "4df42b4df9ca7f17",
+   "outputs": [],
+   "execution_count": 43
+  },
+  {
+   "metadata": {
+    "ExecuteTime": {
+     "end_time": "2024-08-28T07:44:23.257797Z",
+     "start_time": "2024-08-28T07:44:22.988904Z"
+    }
+   },
+   "cell_type": "code",
+   "source": [
+    "# largest surface to specific label\n",
+    "score_threshold = 0.5\n",
+    "folder = 'rayure'\n",
+    "for root, dirs, files in os.walk(os.path.join(root_dataset_folder, 'pre-annotations', folder)):\n",
+    "    for a_file in files:\n",
+    "        if a_file.endswith('.json'):\n",
+    "            filepath = os.path.join(root, a_file)\n",
+    "            with open(filepath, 'r') as f:\n",
+    "                file_content_json = json.load(f)\n",
+    "\n",
+    "            predictions = file_content_json['predictions']\n",
+    "            annotations = predictions[0]['result']\n",
+    "            annotations = list(filter(lambda x: x[\"score\"] > 0.5, annotations))\n",
+    "            if len(annotations) > 0:               \n",
+    "                only_annotation = max(annotations, key=lambda p: p['value']['width'] * p['value']['height'])\n",
+    "                only_annotation['value']['rectanglelabels'] = [folder]\n",
+    "                result = [only_annotation]\n",
+    "            else:\n",
+    "                result = []\n",
+    "\n",
+    "            with open(filepath, 'w') as f:\n",
+    "                json.dump({\n",
+    "                    **file_content_json,\n",
+    "                    'predictions': [\n",
+    "                        {\n",
+    "                            **predictions[0],\n",
+    "                            \"result\": result\n",
+    "                        }\n",
+    "                    ]\n",
+    "                }, f)"
+   ],
+   "id": "bba8c29714c2ab7b",
+   "outputs": [],
+   "execution_count": 21
+  },
+  {
+   "metadata": {},
+   "cell_type": "code",
+   "outputs": [],
+   "execution_count": null,
+   "source": "",
+   "id": "262fe6ef762f57d1"
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 2
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython2",
+   "version": "2.7.6"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}

yolo8.ipynb

@@ -0,0 +1,101 @@
+{
+ "cells": [
+  {
+   "metadata": {},
+   "cell_type": "code",
+   "source": [
+    "from ultralytics import YOLO\n",
+    "\n",
+    "original_model_name = 'yolov8s'\n",
+    "original_model_nickname = 'fashionpedia'\n",
+    "yolo_model = YOLO(original_model_name)"
+   ],
+   "id": "688b209f5c7b833e",
+   "outputs": [],
+   "execution_count": null
+  },
+  {
+   "metadata": {},
+   "cell_type": "code",
+   "source": [
+    "results = yolo_model.train(data='fashionpedia.yaml', epochs=300, batch=32, save_period=1, plots=True,\n",
+    "                           name=f'{original_model_name}-{original_model_nickname}', exist_ok=True, resume=False,\n",
+    "                           deterministic=False, close_mosaic=50, val=False,\n",
+    "                           label_smoothing=0.05, workers=20, cache=False, task=\"detect\",\n",
+    "                           translate=0.2, degrees=30, shear=10, mosaic=0.8, perspective=0.000_01\n",
+    "                           )"
+   ],
+   "id": "dd14c466280421c4",
+   "outputs": [],
+   "execution_count": null
+  },
+  {
+   "metadata": {},
+   "cell_type": "code",
+   "source": [],
+   "id": "8884f7e383561fc2",
+   "outputs": [],
+   "execution_count": null
+  },
+  {
+   "metadata": {},
+   "cell_type": "code",
+   "source": [
+    "# load saved model\n",
+    "yolo_model_loaded = YOLO(\"shared/yolov8n-fashionpedia-1.torchscript\", task='detect')"
+   ],
+   "id": "f427e4ba2cff784",
+   "outputs": [],
+   "execution_count": null
+  },
+  {
+   "metadata": {},
+   "cell_type": "code",
+   "source": [
+    "# Run batched inference on a list of images\n",
+    "inference_results = yolo_model_loaded(['./test/courte.png',\n",
+    "                                       './test/longue.png',\n",
+    "                                       './test/rayure.png',\n",
+    "                                       './test/sans-manche.png',\n",
+    "                                       './test/trois-quart.png'\n",
+    "                                       ], verbose=False)\n",
+    "for result in inference_results:\n",
+    "    boxes = result.boxes  # Boxes object for bounding box outputs\n",
+    "    probs = result.probs  # Probs object for classification outputs\n",
+    "    result.show()  # display to screen"
+   ],
+   "id": "f0815af787b52825",
+   "outputs": [],
+   "execution_count": null
+  },
+  {
+   "metadata": {},
+   "cell_type": "code",
+   "source": [],
+   "id": "793f625e9139f76d",
+   "outputs": [],
+   "execution_count": null
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 2
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython2",
+   "version": "2.7.6"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}