mirror of
https://github.com/deepinsight/insightface.git
synced 2025-12-30 08:02:27 +00:00
536 lines
65 KiB
Plaintext
536 lines
65 KiB
Plaintext
{
|
||
"cells": [
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 1,
|
||
"metadata": {},
|
||
"outputs": [
|
||
{
|
||
"name": "stderr",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"/home/jd4615/miniconda3/envs/insightface/lib/python2.7/site-packages/sklearn/utils/fixes.py:313: FutureWarning: numpy not_equal will not check object identity in the future. The comparison did not return the same result as suggested by the identity (`is`)) and will change.\n",
|
||
" _nan_object_mask = _nan_object_array != _nan_object_array\n"
|
||
]
|
||
}
|
||
],
|
||
"source": [
|
||
"import os\n",
|
||
"import numpy as np\n",
|
||
"import cPickle\n",
|
||
"from sklearn.metrics import roc_curve, auc\n",
|
||
"import matplotlib.pyplot as plt\n",
|
||
"import timeit\n",
|
||
"import sklearn\n",
|
||
"import cv2\n",
|
||
"import sys\n",
|
||
"import glob\n",
|
||
"sys.path.append('./recognition')\n",
|
||
"from embedding import Embedding\n",
|
||
"from menpo.visualize import print_progress\n",
|
||
"from menpo.visualize.viewmatplotlib import sample_colours_from_colourmap\n",
|
||
"from prettytable import PrettyTable\n",
|
||
"from pathlib import Path\n",
|
||
"import warnings \n",
|
||
"warnings.filterwarnings(\"ignore\") "
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 2,
|
||
"metadata": {},
|
||
"outputs": [],
|
||
"source": [
|
||
"def read_template_media_list(path):\n",
|
||
" ijb_meta = np.loadtxt(path, dtype=str)\n",
|
||
" templates = ijb_meta[:,1].astype(np.int)\n",
|
||
" medias = ijb_meta[:,2].astype(np.int)\n",
|
||
" return templates, medias"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 4,
|
||
"metadata": {},
|
||
"outputs": [],
|
||
"source": [
|
||
"def read_template_pair_list(path):\n",
|
||
" pairs = np.loadtxt(path, dtype=str)\n",
|
||
" t1 = pairs[:,0].astype(np.int)\n",
|
||
" t2 = pairs[:,1].astype(np.int)\n",
|
||
" label = pairs[:,2].astype(np.int)\n",
|
||
" return t1, t2, label"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 4,
|
||
"metadata": {},
|
||
"outputs": [],
|
||
"source": [
|
||
"def read_image_feature(path):\n",
|
||
" with open(path, 'rb') as fid:\n",
|
||
" img_feats = cPickle.load(fid)\n",
|
||
" return img_feats"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 5,
|
||
"metadata": {},
|
||
"outputs": [],
|
||
"source": [
|
||
"def get_image_feature(img_path, img_list_path, model_path, gpu_id):\n",
|
||
" img_list = open(img_list_path)\n",
|
||
" embedding = Embedding(model_path, 0, gpu_id)\n",
|
||
" files = img_list.readlines()\n",
|
||
" img_feats = []\n",
|
||
" faceness_scores = []\n",
|
||
" for img_index, each_line in enumerate(print_progress(files)):\n",
|
||
" name_lmk_score = each_line.strip().split(' ')\n",
|
||
" img_name = os.path.join(img_path, name_lmk_score[0])\n",
|
||
" img = cv2.imread(img_name)\n",
|
||
" lmk = np.array([float(x) for x in name_lmk_score[1:-1]], dtype=np.float32)\n",
|
||
" lmk = lmk.reshape( (5,2) )\n",
|
||
" img_feats.append(embedding.get(img,lmk))\n",
|
||
" faceness_scores.append(name_lmk_score[-1])\n",
|
||
" img_feats = np.array(img_feats).astype(np.float32)\n",
|
||
" faceness_scores = np.array(faceness_scores).astype(np.float32)\n",
|
||
" return img_feats, faceness_scores"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 6,
|
||
"metadata": {},
|
||
"outputs": [],
|
||
"source": [
|
||
"def image2template_feature(img_feats = None, templates = None, medias = None):\n",
|
||
" # ==========================================================\n",
|
||
" # 1. face image feature l2 normalization. img_feats:[number_image x feats_dim]\n",
|
||
" # 2. compute media feature.\n",
|
||
" # 3. compute template feature.\n",
|
||
" # ========================================================== \n",
|
||
" unique_templates = np.unique(templates)\n",
|
||
" template_feats = np.zeros((len(unique_templates), img_feats.shape[1]))\n",
|
||
"\n",
|
||
" for count_template, uqt in enumerate(unique_templates):\n",
|
||
" (ind_t,) = np.where(templates == uqt)\n",
|
||
" face_norm_feats = img_feats[ind_t]\n",
|
||
" face_medias = medias[ind_t]\n",
|
||
" unique_medias, unique_media_counts = np.unique(face_medias, return_counts=True)\n",
|
||
" media_norm_feats = []\n",
|
||
" for u,ct in zip(unique_medias, unique_media_counts):\n",
|
||
" (ind_m,) = np.where(face_medias == u)\n",
|
||
" if ct == 1:\n",
|
||
" media_norm_feats += [face_norm_feats[ind_m]]\n",
|
||
" else: # image features from the same video will be aggregated into one feature\n",
|
||
" media_norm_feats += [np.mean(face_norm_feats[ind_m], 0, keepdims=True)]\n",
|
||
" media_norm_feats = np.array(media_norm_feats)\n",
|
||
" # media_norm_feats = media_norm_feats / np.sqrt(np.sum(media_norm_feats ** 2, -1, keepdims=True))\n",
|
||
" template_feats[count_template] = np.sum(media_norm_feats, 0)\n",
|
||
" if count_template % 2000 == 0: \n",
|
||
" print('Finish Calculating {} template features.'.format(count_template))\n",
|
||
" template_norm_feats = template_feats / np.sqrt(np.sum(template_feats ** 2, -1, keepdims=True))\n",
|
||
" return template_norm_feats, unique_templates"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 7,
|
||
"metadata": {},
|
||
"outputs": [],
|
||
"source": [
|
||
"def verification(template_norm_feats = None, unique_templates = None, p1 = None, p2 = None):\n",
|
||
" # ==========================================================\n",
|
||
" # Compute set-to-set Similarity Score.\n",
|
||
" # ==========================================================\n",
|
||
" template2id = np.zeros((max(unique_templates)+1,1),dtype=int)\n",
|
||
" for count_template, uqt in enumerate(unique_templates):\n",
|
||
" template2id[uqt] = count_template\n",
|
||
" \n",
|
||
" score = np.zeros((len(p1),)) # save cosine distance between pairs \n",
|
||
"\n",
|
||
" total_pairs = np.array(range(len(p1)))\n",
|
||
" batchsize = 100000 # small batchsize instead of all pairs in one batch due to the memory limiation\n",
|
||
" sublists = [total_pairs[i:i + batchsize] for i in range(0, len(p1), batchsize)]\n",
|
||
" total_sublists = len(sublists)\n",
|
||
" for c, s in enumerate(sublists):\n",
|
||
" feat1 = template_norm_feats[template2id[p1[s]]]\n",
|
||
" feat2 = template_norm_feats[template2id[p2[s]]]\n",
|
||
" similarity_score = np.sum(feat1 * feat2, -1)\n",
|
||
" score[s] = similarity_score.flatten()\n",
|
||
" if c % 10 == 0:\n",
|
||
" print('Finish {}/{} pairs.'.format(c, total_sublists))\n",
|
||
" return score"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 8,
|
||
"metadata": {},
|
||
"outputs": [],
|
||
"source": [
|
||
"def read_score(path):\n",
|
||
" with open(path, 'rb') as fid:\n",
|
||
" img_feats = cPickle.load(fid)\n",
|
||
" return img_feats"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "markdown",
|
||
"metadata": {},
|
||
"source": [
|
||
"# Step1: Load Meta Data"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 9,
|
||
"metadata": {},
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"Time: 0.83 s. \n"
|
||
]
|
||
}
|
||
],
|
||
"source": [
|
||
"# =============================================================\n",
|
||
"# load image and template relationships for template feature embedding\n",
|
||
"# tid --> template id, mid --> media id \n",
|
||
"# format:\n",
|
||
"# image_name tid mid\n",
|
||
"# =============================================================\n",
|
||
"start = timeit.default_timer()\n",
|
||
"templates, medias = read_template_media_list(os.path.join('IJBB/meta', 'ijbb_face_tid_mid.txt'))\n",
|
||
"stop = timeit.default_timer()\n",
|
||
"print('Time: %.2f s. ' % (stop - start))"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 5,
|
||
"metadata": {},
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"Time: 31.75 s. \n"
|
||
]
|
||
}
|
||
],
|
||
"source": [
|
||
"# =============================================================\n",
|
||
"# load template pairs for template-to-template verification\n",
|
||
"# tid : template id, label : 1/0\n",
|
||
"# format:\n",
|
||
"# tid_1 tid_2 label\n",
|
||
"# =============================================================\n",
|
||
"start = timeit.default_timer()\n",
|
||
"p1, p2, label = read_template_pair_list(os.path.join('IJBB/meta', 'ijbb_template_pair_label.txt'))\n",
|
||
"stop = timeit.default_timer()\n",
|
||
"print('Time: %.2f s. ' % (stop - start))"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "markdown",
|
||
"metadata": {},
|
||
"source": [
|
||
"# Step 2: Get Image Features"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 11,
|
||
"metadata": {},
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"('loading', './pretrained_models/VGG2-ResNet50-Arcface/model', 0)\n",
|
||
"[====================] 100% (227630/227630) - done. \n",
|
||
"Time: 2386.28 s. \n",
|
||
"Feature Shape: (227630 , 1024) .\n"
|
||
]
|
||
}
|
||
],
|
||
"source": [
|
||
"# =============================================================\n",
|
||
"# load image features \n",
|
||
"# format:\n",
|
||
"# img_feats: [image_num x feats_dim] (227630, 512)\n",
|
||
"# =============================================================\n",
|
||
"start = timeit.default_timer()\n",
|
||
"#img_feats = read_image_feature('./MS1MV2/IJBB_MS1MV2_r100_arcface.pkl')\n",
|
||
"img_path = './IJBB/loose_crop'\n",
|
||
"img_list_path = './IJBB/meta/ijbb_name_5pts_score.txt'\n",
|
||
"model_path = './pretrained_models/VGG2-ResNet50-Arcface/model'\n",
|
||
"gpu_id = 0\n",
|
||
"img_feats, faceness_scores = get_image_feature(img_path, img_list_path, model_path, gpu_id)\n",
|
||
"stop = timeit.default_timer()\n",
|
||
"print('Time: %.2f s. ' % (stop - start))\n",
|
||
"print('Feature Shape: ({} , {}) .'.format(img_feats.shape[0], img_feats.shape[1]))"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "markdown",
|
||
"metadata": {},
|
||
"source": [
|
||
"# Step3: Get Template Features"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 12,
|
||
"metadata": {},
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"Finish Calculating 0 template features.\n",
|
||
"Finish Calculating 2000 template features.\n",
|
||
"Finish Calculating 4000 template features.\n",
|
||
"Finish Calculating 6000 template features.\n",
|
||
"Finish Calculating 8000 template features.\n",
|
||
"Finish Calculating 10000 template features.\n",
|
||
"Finish Calculating 12000 template features.\n",
|
||
"Time: 3.41 s. \n"
|
||
]
|
||
}
|
||
],
|
||
"source": [
|
||
"# =============================================================\n",
|
||
"# compute template features from image features.\n",
|
||
"# =============================================================\n",
|
||
"start = timeit.default_timer()\n",
|
||
"# ========================================================== \n",
|
||
"# Norm feature before aggregation into template feature?\n",
|
||
"# Feature norm from embedding network and faceness score are able to decrease weights for noise samples (not face).\n",
|
||
"# ========================================================== \n",
|
||
"# 1. FaceScore （Feature Norm）\n",
|
||
"# 2. FaceScore （Detector）\n",
|
||
"\n",
|
||
"use_norm_score = False # if True, TestMode(N1) \n",
|
||
"use_detector_score = True # if True, TestMode(D1)\n",
|
||
"use_flip_test = True # if True, TestMode(F1)\n",
|
||
"\n",
|
||
"if use_flip_test:\n",
|
||
" # concat --- F1\n",
|
||
" #img_input_feats = img_feats \n",
|
||
" # add --- F2\n",
|
||
" img_input_feats = img_feats[:,0:img_feats.shape[1]/2] + img_feats[:,img_feats.shape[1]/2:]\n",
|
||
"else:\n",
|
||
" img_input_feats = img_feats[:,0:img_feats.shape[1]/2]\n",
|
||
" \n",
|
||
"if use_norm_score:\n",
|
||
" img_input_feats = img_input_feats\n",
|
||
"else:\n",
|
||
" # normalise features to remove norm information\n",
|
||
" img_input_feats = img_input_feats / np.sqrt(np.sum(img_input_feats ** 2, -1, keepdims=True)) \n",
|
||
" \n",
|
||
"if use_detector_score:\n",
|
||
" img_input_feats = img_input_feats * np.matlib.repmat(faceness_scores[:,np.newaxis], 1, img_input_feats.shape[1])\n",
|
||
"else:\n",
|
||
" img_input_feats = img_input_feats\n",
|
||
"\n",
|
||
"template_norm_feats, unique_templates = image2template_feature(img_input_feats, templates, medias)\n",
|
||
"stop = timeit.default_timer()\n",
|
||
"print('Time: %.2f s. ' % (stop - start))"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "markdown",
|
||
"metadata": {},
|
||
"source": [
|
||
"# Step 4: Get Template Similarity Scores"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 13,
|
||
"metadata": {},
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"Finish 0/81 pairs.\n",
|
||
"Finish 10/81 pairs.\n",
|
||
"Finish 20/81 pairs.\n",
|
||
"Finish 30/81 pairs.\n",
|
||
"Finish 40/81 pairs.\n",
|
||
"Finish 50/81 pairs.\n",
|
||
"Finish 60/81 pairs.\n",
|
||
"Finish 70/81 pairs.\n",
|
||
"Finish 80/81 pairs.\n",
|
||
"Time: 38.38 s. \n"
|
||
]
|
||
}
|
||
],
|
||
"source": [
|
||
"# =============================================================\n",
|
||
"# compute verification scores between template pairs.\n",
|
||
"# =============================================================\n",
|
||
"start = timeit.default_timer()\n",
|
||
"score = verification(template_norm_feats, unique_templates, p1, p2)\n",
|
||
"stop = timeit.default_timer()\n",
|
||
"print('Time: %.2f s. ' % (stop - start))"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 14,
|
||
"metadata": {},
|
||
"outputs": [],
|
||
"source": [
|
||
"score_save_name = './IJBB/result/VGG2-ResNet50-ArcFace-TestMode(N0D1F2).npy'\n",
|
||
"np.save(score_save_name, score)"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "markdown",
|
||
"metadata": {},
|
||
"source": [
|
||
"# Step 5: Get ROC Curves and TPR@FPR Table"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 6,
|
||
"metadata": {},
|
||
"outputs": [
|
||
{
|
||
"data": {
|
||
"image/png": "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\n",
|
||
"text/plain": [
|
||
"<Figure size 432x288 with 1 Axes>"
|
||
]
|
||
},
|
||
"metadata": {
|
||
"needs_background": "light"
|
||
},
|
||
"output_type": "display_data"
|
||
}
|
||
],
|
||
"source": [
|
||
"score_save_path = './IJBB/result'\n",
|
||
"files = glob.glob(score_save_path + '/VGG2*.npy') \n",
|
||
"methods = []\n",
|
||
"scores = []\n",
|
||
"for file in files:\n",
|
||
" methods.append(Path(file).stem)\n",
|
||
" scores.append(np.load(file)) \n",
|
||
"methods = np.array(methods)\n",
|
||
"scores = dict(zip(methods,scores))\n",
|
||
"colours = dict(zip(methods, sample_colours_from_colourmap(methods.shape[0], 'Set2')))\n",
|
||
"#x_labels = [1/(10**x) for x in np.linspace(6, 0, 6)]\n",
|
||
"x_labels = [10**-6, 10**-5, 10**-4,10**-3, 10**-2, 10**-1]\n",
|
||
"tpr_fpr_table = PrettyTable(['Methods'] + map(str, x_labels))\n",
|
||
"fig = plt.figure()\n",
|
||
"for method in methods:\n",
|
||
" fpr, tpr, _ = roc_curve(label, scores[method])\n",
|
||
" roc_auc = auc(fpr, tpr)\n",
|
||
" fpr = np.flipud(fpr)\n",
|
||
" tpr = np.flipud(tpr) # select largest tpr at same fpr\n",
|
||
" plt.plot(fpr, tpr, color=colours[method], lw=1, label=('[%s (AUC = %0.4f %%)]' % (method.split('-')[-1], roc_auc*100)))\n",
|
||
" tpr_fpr_row = []\n",
|
||
" tpr_fpr_row.append(method)\n",
|
||
" for fpr_iter in np.arange(len(x_labels)):\n",
|
||
" _, min_index = min(list(zip(abs(fpr-x_labels[fpr_iter]), range(len(fpr)))))\n",
|
||
" tpr_fpr_row.append('%.4f' % tpr[min_index])\n",
|
||
" tpr_fpr_table.add_row(tpr_fpr_row)\n",
|
||
"plt.xlim([10**-6, 0.1])\n",
|
||
"plt.ylim([0.3, 1.0])\n",
|
||
"plt.grid(linestyle='--', linewidth=1)\n",
|
||
"plt.xticks(x_labels) \n",
|
||
"plt.yticks(np.linspace(0.3, 1.0, 8, endpoint=True)) \n",
|
||
"plt.xscale('log')\n",
|
||
"plt.xlabel('False Positive Rate')\n",
|
||
"plt.ylabel('True Positive Rate')\n",
|
||
"plt.title('ROC on IJB-B')\n",
|
||
"plt.legend(loc=\"lower right\")\n",
|
||
"plt.show()\n",
|
||
"#fig.savefig('IJB-B.pdf')"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": 7,
|
||
"metadata": {},
|
||
"outputs": [
|
||
{
|
||
"name": "stdout",
|
||
"output_type": "stream",
|
||
"text": [
|
||
"+----------------------------------------+--------+--------+--------+--------+--------+--------+\n",
|
||
"| Methods | 1e-06 | 1e-05 | 0.0001 | 0.001 | 0.01 | 0.1 |\n",
|
||
"+----------------------------------------+--------+--------+--------+--------+--------+--------+\n",
|
||
"| VGG2-ResNet50-ArcFace-TestMode(N1D1F2) | 0.4044 | 0.8145 | 0.9056 | 0.9497 | 0.9779 | 0.9922 |\n",
|
||
"| VGG2-ResNet50-ArcFace-TestMode(N1D0F0) | 0.4035 | 0.8038 | 0.8976 | 0.9437 | 0.9755 | 0.9914 |\n",
|
||
"| VGG2-ResNet50-ArcFace-TestMode(N1D1F1) | 0.3940 | 0.8124 | 0.9028 | 0.9479 | 0.9770 | 0.9919 |\n",
|
||
"| VGG2-ResNet50-ArcFace-TestMode(N0D0F0) | 0.3893 | 0.8050 | 0.8990 | 0.9448 | 0.9759 | 0.9918 |\n",
|
||
"| VGG2-ResNet50-ArcFace-TestMode(N1D1F0) | 0.4098 | 0.8123 | 0.9022 | 0.9463 | 0.9766 | 0.9918 |\n",
|
||
"| VGG2-ResNet50-ArcFace-TestMode(N0D1F0) | 0.3949 | 0.8130 | 0.9036 | 0.9471 | 0.9767 | 0.9919 |\n",
|
||
"| VGG2-ResNet50-ArcFace-TestMode(N0D1F2) | 0.4011 | 0.8210 | 0.9069 | 0.9500 | 0.9779 | 0.9924 |\n",
|
||
"+----------------------------------------+--------+--------+--------+--------+--------+--------+\n"
|
||
]
|
||
}
|
||
],
|
||
"source": [
|
||
"print(tpr_fpr_table)"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "code",
|
||
"execution_count": null,
|
||
"metadata": {},
|
||
"outputs": [],
|
||
"source": [
|
||
"# setting N0D1F2 is the best"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "markdown",
|
||
"metadata": {},
|
||
"source": [
|
||
"# Test Setting Conclusions"
|
||
]
|
||
},
|
||
{
|
||
"cell_type": "markdown",
|
||
"metadata": {},
|
||
"source": [
|
||
"#### (1) add is better than concat for the flip test (N1D1F2 v.s. N1D1F1)\n",
|
||
"#### (2) detection score contains some faceness information to decrease weights of noise samples within the template (N0D1F0 v.s. N0D0F0)"
|
||
]
|
||
}
|
||
],
|
||
"metadata": {
|
||
"kernelspec": {
|
||
"display_name": "Python 2",
|
||
"language": "python",
|
||
"name": "python2"
|
||
},
|
||
"language_info": {
|
||
"codemirror_mode": {
|
||
"name": "ipython",
|
||
"version": 2
|
||
},
|
||
"file_extension": ".py",
|
||
"mimetype": "text/x-python",
|
||
"name": "python",
|
||
"nbconvert_exporter": "python",
|
||
"pygments_lexer": "ipython2",
|
||
"version": "2.7.15"
|
||
}
|
||
},
|
||
"nbformat": 4,
|
||
"nbformat_minor": 2
|
||
}
|