@@ -2,25 +2,17 @@
"cells" : [
{
"cell_type" : " code"
"execution_count" : 2 ,
"execution_count" : 101 ,
"metadata" : {},
"outputs" : [
{
"name" : " stderr"
"output_type" : " stream"
"text" : [
" Using TensorFlow backend.\n "
]
}
],
"outputs" : [],
"source" : [
" import sys\n "
" from keras.applications.resnet50 import ResNet50, preprocess_input, decode_predictions\n "
" from keras.preprocessing.image import load_img, img_to_array\n "
" import numpy as np\n "
" \n "
" sys.path.append('pyttr')\n "
" from pyttr import ttrtypes as ttr, utils as ttru "
" from pyttr.ttrtypes import BType, PType, Possibility, Pred, RecType, Rec, FunType, Fun "
]
},
{
@@ -32,26 +24,16 @@
},
{
"cell_type" : " code"
"execution_count" : 3 ,
"execution_count" : 49 ,
"metadata" : {},
"outputs" : [
{
"name" : " stdout"
"output_type" : " stream"
"text" : [
" WARNING:tensorflow:From /usr/local/lib/python3.5/dist-packages/keras/backend/tensorflow_backend.py:1264: calling reduce_prod (from tensorflow.python.ops.math_ops) with keep_dims is deprecated and will be removed in a future version.\n "
" Instructions for updating:\n "
" keep_dims is deprecated, use keepdims instead\n "
]
}
],
"outputs" : [],
"source" : [
" model = ResNet50()"
]
},
{
"cell_type" : " code"
"execution_count" : 4 ,
"execution_count" : 50 ,
"metadata" : {},
"outputs" : [
{
@@ -60,7 +42,7 @@
" 'palace'"
]
},
"execution_count" : 4 ,
"execution_count" : 50 ,
"metadata" : {},
"output_type" : " execute_result"
}
@@ -76,24 +58,24 @@
},
{
"cell_type" : " code"
"execution_count" : 5 ,
"execution_count" : 51 ,
"metadata" : {},
"outputs" : [],
"source" : [
" M = ttr. Possibility('M1')\n "
" Ind = ttr. BType('Ind')\n "
" M = Possibility('M1')\n "
" Ind = BType('Ind')\n "
" Ind.in_poss(M)\n "
" \n "
" def see(path):\n "
" pred = ttr. Pred(classify(path), [Ind])\n "
" pred = Pred(classify(path), [Ind])\n "
" name = path.split('/')[-1].split('.')[0]\n "
" Ind.judge(name)\n "
" ttr. PType(pred, [name]).in_poss(M).judge(path)"
" PType(pred, [name]).in_poss(M).judge(path)"
]
},
{
"cell_type" : " code"
"execution_count" : 6 ,
"execution_count" : 52 ,
"metadata" : {},
"outputs" : [
{
@@ -120,7 +102,7 @@
" see('res/dogcar.jpg')\n "
" #see('/home/arildm/Pictures/star.png')\n "
" #see('/home/arildm/Pictures/skocaj.png')\n "
" print(ttru .show(M ))"
" print(M .show())"
]
},
{
@@ -132,7 +114,7 @@
},
{
"cell_type" : " code"
"execution_count" : 20 ,
"execution_count" : 53 ,
"metadata" : {},
"outputs" : [
{
@@ -148,7 +130,7 @@
" words = ['volcano', 'palace', 'dog', 'canal']\n "
" ptypes = []\n "
" for word in words:\n "
" ptype = ttr. PType(ttr. Pred(word, [Ind]), ['x'])\n "
" ptype = PType(Pred(word, [Ind]), ['x'])\n "
" ptype.learn_witness_condition(lambda path, word=word: classify(path) == word)\n "
" ptypes.append(ptype)\n "
" for path in ['res/volcano2.jpg', 'res/canal.jpg', 'res/dogcar.jpg']:\n "
@@ -168,163 +150,32 @@
},
{
"cell_type" : " code"
"execution_count" : 21 ,
"execution_count" : 70 ,
"metadata" : {},
"outputs" : [
{
"name" : " stdout"
"output_type" : " stream"
"text" : [
" ['{pfun : lambda x:Image . volcano(x), reg : <PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=1200x900 at 0x7FCB86B85080 >}']\n "
" ['{pfun : lambda x:Image . volcano(x), reg : <PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=1200x900 at 0x7FCB0B03CEF0 >}']\n "
]
}
],
"source" : [
" imageT = ttr. BType('Image')\n "
" imageT = BType('Image')\n "
" \n "
" def odf (path):\n "
" def object_detector_simple (path):\n "
" det = set()\n "
" for ptype in ptypes:\n "
" if ptype.query(path):\n "
" det.add(ttr. RecType({\n "
" det.add(RecType({\n "
" 'reg': str(load_img(path)),\n "
" 'pfun': ttr. Fun('x', imageT, ptype),\n "
" 'pfun': Fun('x', imageT, ptype),\n "
" }))\n "
" return det\n "
" \n "
" print([t.show() for t in odf('res/volcano2.jpg')])"
]
},
{
"cell_type" : " markdown"
"metadata" : {},
"source" : [
" ## YOLO\n "
" \n "
" Enter object detection.\n "
" \n "
" Requires OpenCV and [Darkflow](https://github.com/thtrieu/darkflow). yolo.weights is from [Yolo](https://pjreddie.com/darknet/yolo/)."
]
},
{
"cell_type" : " code"
"execution_count" : 45 ,
"metadata" : {},
"outputs" : [
{
"name" : " stderr"
"output_type" : " stream"
"text" : [
" /opt/darkflow/darkflow/dark/darknet.py:54: UserWarning: ./cfg/yolo.cfg not found, use yolo/yolo.cfg instead\n "
" cfg_path, FLAGS.model))\n "
]
},
{
"name" : " stdout"
"output_type" : " stream"
"text" : [
" Parsing yolo/yolo.cfg\n "
" Loading yolo/yolo.weights ...\n "
" Successfully identified 203934260 bytes\n "
" Finished in 0.014590978622436523s\n "
" Model has a coco model name, loading coco labels.\n "
" \n "
" Building net ...\n "
" Source | Train? | Layer description | Output size\n "
" -------+--------+----------------------------------+---------------\n "
" | | input | (?, 608, 608, 3)\n "
" Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 608, 608, 32)\n "
" Load | Yep! | maxp 2x2p0_2 | (?, 304, 304, 32)\n "
" Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 304, 304, 64)\n "
" Load | Yep! | maxp 2x2p0_2 | (?, 152, 152, 64)\n "
" Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 152, 152, 128)\n "
" Load | Yep! | conv 1x1p0_1 +bnorm leaky | (?, 152, 152, 64)\n "
" Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 152, 152, 128)\n "
" Load | Yep! | maxp 2x2p0_2 | (?, 76, 76, 128)\n "
" Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 76, 76, 256)\n "
" Load | Yep! | conv 1x1p0_1 +bnorm leaky | (?, 76, 76, 128)\n "
" Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 76, 76, 256)\n "
" Load | Yep! | maxp 2x2p0_2 | (?, 38, 38, 256)\n "
" Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 38, 38, 512)\n "
" Load | Yep! | conv 1x1p0_1 +bnorm leaky | (?, 38, 38, 256)\n "
" Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 38, 38, 512)\n "
" Load | Yep! | conv 1x1p0_1 +bnorm leaky | (?, 38, 38, 256)\n "
" Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 38, 38, 512)\n "
" Load | Yep! | maxp 2x2p0_2 | (?, 19, 19, 512)\n "
" Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 19, 19, 1024)\n "
" Load | Yep! | conv 1x1p0_1 +bnorm leaky | (?, 19, 19, 512)\n "
" Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 19, 19, 1024)\n "
" Load | Yep! | conv 1x1p0_1 +bnorm leaky | (?, 19, 19, 512)\n "
" Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 19, 19, 1024)\n "
" Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 19, 19, 1024)\n "
" Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 19, 19, 1024)\n "
" Load | Yep! | concat [16] | (?, 38, 38, 512)\n "
" Load | Yep! | conv 1x1p0_1 +bnorm leaky | (?, 38, 38, 64)\n "
" Load | Yep! | local flatten 2x2 | (?, 19, 19, 256)\n "
" Load | Yep! | concat [27, 24] | (?, 19, 19, 1280)\n "
" Load | Yep! | conv 3x3p1_1 +bnorm leaky | (?, 19, 19, 1024)\n "
" Load | Yep! | conv 1x1p0_1 linear | (?, 19, 19, 425)\n "
" -------+--------+----------------------------------+---------------\n "
" Running entirely on CPU\n "
" Finished in 7.629829406738281s\n "
" \n "
]
}
],
"source" : [
" from darkflow.net.build import TFNet\n "
" import cv2\n "
" \n "
" options = {\" model\" : \" yolo/yolo.cfg\" , \" load\" : \" yolo/yolo.weights\" , \" threshold\" : 0.1}\n "
" \n "
" tfnet = TFNet(options)"
]
},
{
"cell_type" : " code"
"execution_count" : 42 ,
"metadata" : {},
"outputs" : [],
"source" : [
" def object_detector(cv_image):\n "
" s = set()\n "
" for o in tfnet.return_predict(imgcv):\n "
" if o['confidence'] > 0.1:\n "
" ptype = ttr.PType(ttr.Pred(o['label'], [Ind]), ['x'])\n "
" pfun = ttr.Fun('x', imageT, ptype)\n "
" s.add(ttr.RecType({'reg': (o['topleft'], o['bottomright']), 'pfun': pfun}))\n "
" return s"
]
},
{
"cell_type" : " code"
"execution_count" : 43 ,
"metadata" : {},
"outputs" : [
{
"name" : " stdout"
"output_type" : " stream"
"text" : [
" {pfun : lambda x:Image . car(x), reg : ({x = 170, y = 89}, {x = 1019, y = 1006})}\n "
" {pfun : lambda x:Image . chair(x), reg : ({x = 191, y = 308}, {x = 905, y = 1079})}\n "
" {pfun : lambda x:Image . person(x), reg : ({x = 356, y = 236}, {x = 1065, y = 1072})}\n "
" {pfun : lambda x:Image . dog(x), reg : ({x = 241, y = 320}, {x = 968, y = 1079})}\n "
" {pfun : lambda x:Image . dog(x), reg : ({x = 706, y = 574}, {x = 1079, y = 1067})}\n "
" {pfun : lambda x:Image . person(x), reg : ({x = 3, y = 697}, {x = 386, y = 1079})}\n "
]
}
],
"source" : [
" for t in object_detector(cv2.imread('res/dogcar.jpg')):\n "
" print(t.show())"
]
},
{
"cell_type" : " markdown"
"metadata" : {},
"source" : [
" Now what about the witness condition??"
]
}
],
"metadata" : {
