You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
283 lines
603 KiB
283 lines
603 KiB
5 years ago
|
{
|
||
|
|
"cells": [
|
||
|
|
{
|
||
|
|
"cell_type": "code",
|
||
|
|
"execution_count": 1,
|
||
|
|
"metadata": {
|
||
|
|
"collapsed": true,
|
||
|
|
"deletable": true,
|
||
|
|
"editable": true
|
||
|
|
},
|
||
|
|
"outputs": [],
|
||
|
|
"source": [
|
||
|
|
"import os\n",
|
||
|
|
"import math\n",
|
||
|
|
"import random\n",
|
||
|
|
"\n",
|
||
|
|
"import numpy as np\n",
|
||
|
|
"import tensorflow as tf\n",
|
||
|
|
"import cv2\n",
|
||
|
|
"\n",
|
||
|
|
"slim = tf.contrib.slim"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "code",
|
||
|
|
"execution_count": 2,
|
||
|
|
"metadata": {
|
||
|
|
"collapsed": true,
|
||
|
|
"deletable": true,
|
||
|
|
"editable": true
|
||
|
|
},
|
||
|
|
"outputs": [],
|
||
|
|
"source": [
|
||
|
|
"%matplotlib inline\n",
|
||
|
|
"import matplotlib.pyplot as plt\n",
|
||
|
|
"import matplotlib.image as mpimg"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "code",
|
||
|
|
"execution_count": 3,
|
||
|
|
"metadata": {
|
||
|
|
"collapsed": true,
|
||
|
|
"deletable": true,
|
||
|
|
"editable": true
|
||
|
|
},
|
||
|
|
"outputs": [],
|
||
|
|
"source": [
|
||
|
|
"import sys\n",
|
||
|
|
"sys.path.append('../')"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "code",
|
||
|
|
"execution_count": 4,
|
||
|
|
"metadata": {
|
||
|
|
"collapsed": true,
|
||
|
|
"deletable": true,
|
||
|
|
"editable": true
|
||
|
|
},
|
||
|
|
"outputs": [],
|
||
|
|
"source": [
|
||
|
|
"from nets import ssd_vgg_300, ssd_common, np_methods\n",
|
||
|
|
"from preprocessing import ssd_vgg_preprocessing\n",
|
||
|
|
"from notebooks import visualization"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "code",
|
||
|
|
"execution_count": 5,
|
||
|
|
"metadata": {
|
||
|
|
"collapsed": false,
|
||
|
|
"deletable": true,
|
||
|
|
"editable": true
|
||
|
|
},
|
||
|
|
"outputs": [],
|
||
|
|
"source": [
|
||
|
|
"# TensorFlow session: grow memory when needed. TF, DO NOT USE ALL MY GPU MEMORY!!!\n",
|
||
|
|
"gpu_options = tf.GPUOptions(allow_growth=True)\n",
|
||
|
|
"config = tf.ConfigProto(log_device_placement=False, gpu_options=gpu_options)\n",
|
||
|
|
"isess = tf.InteractiveSession(config=config)"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "markdown",
|
||
|
|
"metadata": {
|
||
|
|
"collapsed": false,
|
||
|
|
"deletable": true,
|
||
|
|
"editable": true
|
||
|
|
},
|
||
|
|
"source": [
|
||
|
|
"## SSD 300 Model\n",
|
||
|
|
"\n",
|
||
|
|
"The SSD 300 network takes 300x300 image inputs. In order to feed any image, the latter is resize to this input shape (i.e.`Resize.WARP_RESIZE`). Note that even though it may change the ratio width / height, the SSD model performs well on resized images (and it is the default behaviour in the original Caffe implementation).\n",
|
||
|
|
"\n",
|
||
|
|
"SSD anchors correspond to the default bounding boxes encoded in the network. The SSD net output provides offset on the coordinates and dimensions of these anchors."
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "code",
|
||
|
|
"execution_count": 6,
|
||
|
|
"metadata": {
|
||
|
|
"collapsed": false,
|
||
|
|
"deletable": true,
|
||
|
|
"editable": true,
|
||
|
|
"scrolled": false
|
||
|
|
},
|
||
|
|
"outputs": [],
|
||
|
|
"source": [
|
||
|
|
"# Input placeholder.\n",
|
||
|
|
"net_shape = (300, 300)\n",
|
||
|
|
"data_format = 'NHWC'\n",
|
||
|
|
"img_input = tf.placeholder(tf.uint8, shape=(None, None, 3))\n",
|
||
|
|
"# Evaluation pre-processing: resize to SSD net shape.\n",
|
||
|
|
"image_pre, labels_pre, bboxes_pre, bbox_img = ssd_vgg_preprocessing.preprocess_for_eval(\n",
|
||
|
|
" img_input, None, None, net_shape, data_format, resize=ssd_vgg_preprocessing.Resize.WARP_RESIZE)\n",
|
||
|
|
"image_4d = tf.expand_dims(image_pre, 0)\n",
|
||
|
|
"\n",
|
||
|
|
"# Define the SSD model.\n",
|
||
|
|
"reuse = True if 'ssd_net' in locals() else None\n",
|
||
|
|
"ssd_net = ssd_vgg_300.SSDNet()\n",
|
||
|
|
"with slim.arg_scope(ssd_net.arg_scope(data_format=data_format)):\n",
|
||
|
|
" predictions, localisations, _, _ = ssd_net.net(image_4d, is_training=False, reuse=reuse)\n",
|
||
|
|
"\n",
|
||
|
|
"# Restore SSD model.\n",
|
||
|
|
"ckpt_filename = '../checkpoints/ssd_300_vgg.ckpt'\n",
|
||
|
|
"# ckpt_filename = '../checkpoints/VGG_VOC0712_SSD_300x300_ft_iter_120000.ckpt'\n",
|
||
|
|
"isess.run(tf.global_variables_initializer())\n",
|
||
|
|
"saver = tf.train.Saver()\n",
|
||
|
|
"saver.restore(isess, ckpt_filename)\n",
|
||
|
|
"\n",
|
||
|
|
"# SSD default anchor boxes.\n",
|
||
|
|
"ssd_anchors = ssd_net.anchors(net_shape)"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "markdown",
|
||
|
|
"metadata": {
|
||
|
|
"deletable": true,
|
||
|
|
"editable": true
|
||
|
|
},
|
||
|
|
"source": [
|
||
|
|
"## Post-processing pipeline\n",
|
||
|
|
"\n",
|
||
|
|
"The SSD outputs need to be post-processed to provide proper detections. Namely, we follow these common steps:\n",
|
||
|
|
"\n",
|
||
|
|
"* Select boxes above a classification threshold;\n",
|
||
|
|
"* Clip boxes to the image shape;\n",
|
||
|
|
"* Apply the Non-Maximum-Selection algorithm: fuse together boxes whose Jaccard score > threshold;\n",
|
||
|
|
"* If necessary, resize bounding boxes to original image shape."
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "code",
|
||
|
|
"execution_count": 7,
|
||
|
|
"metadata": {
|
||
|
|
"collapsed": false,
|
||
|
|
"deletable": true,
|
||
|
|
"editable": true,
|
||
|
|
"scrolled": true
|
||
|
|
},
|
||
|
|
"outputs": [],
|
||
|
|
"source": [
|
||
|
|
"# Main image processing routine.\n",
|
||
|
|
"def process_image(img, select_threshold=0.5, nms_threshold=.45, net_shape=(300, 300)):\n",
|
||
|
|
" # Run SSD network.\n",
|
||
|
|
" rimg, rpredictions, rlocalisations, rbbox_img = isess.run([image_4d, predictions, localisations, bbox_img],\n",
|
||
|
|
" feed_dict={img_input: img})\n",
|
||
|
|
" \n",
|
||
|
|
" # Get classes and bboxes from the net outputs.\n",
|
||
|
|
" rclasses, rscores, rbboxes = np_methods.ssd_bboxes_select(\n",
|
||
|
|
" rpredictions, rlocalisations, ssd_anchors,\n",
|
||
|
|
" select_threshold=select_threshold, img_shape=net_shape, num_classes=21, decode=True)\n",
|
||
|
|
" \n",
|
||
|
|
" rbboxes = np_methods.bboxes_clip(rbbox_img, rbboxes)\n",
|
||
|
|
" rclasses, rscores, rbboxes = np_methods.bboxes_sort(rclasses, rscores, rbboxes, top_k=400)\n",
|
||
|
|
" rclasses, rscores, rbboxes = np_methods.bboxes_nms(rclasses, rscores, rbboxes, nms_threshold=nms_threshold)\n",
|
||
|
|
" # Resize bboxes to original image shape. Note: useless for Resize.WARP!\n",
|
||
|
|
" rbboxes = np_methods.bboxes_resize(rbbox_img, rbboxes)\n",
|
||
|
|
" return rclasses, rscores, rbboxes"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "code",
|
||
|
|
"execution_count": 21,
|
||
|
|
"metadata": {
|
||
|
|
"collapsed": false,
|
||
|
|
"deletable": true,
|
||
|
|
"editable": true,
|
||
|
|
"scrolled": false
|
||
|
|
},
|
||
|
|
"outputs": [
|
||
|
|
{
|
||
|
|
"data": {
|
||
|
|
"image/png": "iVBORw0KGgoAAAANSUhEUgAAAloAAAHFCAYAAAAqg1fhAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nrxJETaqVBCe1cbRFXnn
|
||
|
|
"text/plain": [
|
||
|
|
"<matplotlib.figure.Figure at 0x7f349ab800f0>"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
"metadata": {},
|
||
|
|
"output_type": "display_data"
|
||
|
|
}
|
||
|
|
],
|
||
|
|
"source": [
|
||
|
|
"# Test on some demo image and visualize output.\n",
|
||
|
|
"path = '../demo/'\n",
|
||
|
|
"image_names = sorted(os.listdir(path))\n",
|
||
|
|
"\n",
|
||
|
|
"img = mpimg.imread(path + image_names[-5])\n",
|
||
|
|
"rclasses, rscores, rbboxes = process_image(img)\n",
|
||
|
|
"\n",
|
||
|
|
"# visualization.bboxes_draw_on_img(img, rclasses, rscores, rbboxes, visualization.colors_plasma)\n",
|
||
|
|
"visualization.plt_bboxes(img, rclasses, rscores, rbboxes)"
|
||
|
|
]
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "code",
|
||
|
|
"execution_count": null,
|
||
|
|
"metadata": {
|
||
|
|
"collapsed": false,
|
||
|
|
"deletable": true,
|
||
|
|
"editable": true
|
||
|
|
},
|
||
|
|
"outputs": [],
|
||
|
|
"source": []
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "code",
|
||
|
|
"execution_count": null,
|
||
|
|
"metadata": {
|
||
|
|
"collapsed": false,
|
||
|
|
"deletable": true,
|
||
|
|
"editable": true
|
||
|
|
},
|
||
|
|
"outputs": [],
|
||
|
|
"source": []
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "code",
|
||
|
|
"execution_count": null,
|
||
|
|
"metadata": {
|
||
|
|
"collapsed": true,
|
||
|
|
"deletable": true,
|
||
|
|
"editable": true
|
||
|
|
},
|
||
|
|
"outputs": [],
|
||
|
|
"source": []
|
||
|
|
},
|
||
|
|
{
|
||
|
|
"cell_type": "code",
|
||
|
|
"execution_count": null,
|
||
|
|
"metadata": {
|
||
|
|
"collapsed": true,
|
||
|
|
"deletable": true,
|
||
|
|
"editable": true
|
||
|
|
},
|
||
|
|
"outputs": [],
|
||
|
|
"source": []
|
||
|
|
}
|
||
|
|
],
|
||
|
|
"metadata": {
|
||
|
|
"kernelspec": {
|
||
|
|
"display_name": "Python [default]",
|
||
|
|
"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": "3.5.3"
|
||
|
|
}
|
||
|
|
},
|
||
|
|
"nbformat": 4,
|
||
|
|
"nbformat_minor": 2
|
||
|
|
}
|