Skip to content

Commit

Permalink
Fix a bug when generatig the output from the labels
Browse files Browse the repository at this point in the history
  • Loading branch information
@albertomontesg
albertomontesg committed Mar 23, 2017
1 parent 32581d4 commit 72bcb7f
Show file tree
Hide file tree
Showing 3 changed files with 158 additions and 26 deletions.
170 changes: 150 additions & 20 deletions notebooks/15 Test Generation Output.ipynb
View file
Original file line number Diff line number Diff line change
Expand Up @@ -3,39 +3,161 @@
{
"cell_type": "code",
"execution_count": 1,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"import json\n",
"import numpy as np\n",
"from ..src.data"
]
},
{
"cell_type": "code",
"execution_count": 29,
"metadata": {
"collapsed": true
},
"outputs": [],
"source": [
"def import_labels(f):\n",
" ''' Read from a file all the labels from it '''\n",
" lines = f.readlines()\n",
" labels = []\n",
" i = 0\n",
" for l in lines:\n",
" t = l.split('\\t')\n",
" assert int(t[0]) == i\n",
" label = t[1].split('\\n')[0]\n",
" labels.append(label)\n",
" i += 1\n",
" return labels\n",
"\n",
"def to_categorical(y, nb_classes=None):\n",
" ''' Convert class vector (integers from 0 to nb_classes)\n",
" to binary class matrix, for use with categorical_crossentropy.\n",
" '''\n",
" if not nb_classes:\n",
" nb_classes = np.max(y)+1\n",
" Y = np.zeros((len(y), nb_classes))\n",
" for i in range(len(y)):\n",
" Y[i, y[i]] = 1.\n",
" return Y\n",
"\n",
"def generate_output(video_info, labels, length=16):\n",
" ''' Given the info of the vide, generate a vector of classes corresponding the output for each\n",
" clip of the video which features have been extracted.\n",
" '''\n",
" nb_frames = video_info['num_frames']\n",
" last_first_name = nb_frames - length + 1\n",
"\n",
" start_frames = range(0, last_first_name, length)\n",
"\n",
" # Check the output for each frame of the video\n",
" outputs = ['none'] * nb_frames\n",
" for i in range(nb_frames):\n",
" # Pass frame to temporal scale\n",
" t = i / float(nb_frames) * video_info['duration']\n",
" for annotation in video_info['annotations']:\n",
" if t >= annotation['segment'][0] and t <= annotation['segment'][1]:\n",
" outputs[i] = annotation['label']\n",
" label = annotation['label']\n",
" break\n",
"\n",
" instances = []\n",
" for start_frame in start_frames:\n",
" # Obtain the label for this isntance and then its output\n",
" output = None\n",
"\n",
" outs = outputs[start_frame:start_frame+length]\n",
" if outs.count(label) >= length / 2:\n",
" output = labels.index(label)\n",
" else:\n",
" output = 0\n",
" instances.append(output)\n",
"\n",
" return instances"
]
},
{
"cell_type": "code",
"execution_count": 5,
"metadata": {
"collapsed": false
},
"outputs": [],
"source": [
"with open(\"../dataset/labels.txt\", \"r\") as f:\n",
" labels = import_labels(f)"
]
},
{
"cell_type": "code",
"execution_count": 10,
"metadata": {
"collapsed": false
},
"outputs": [
{
"name": "stderr",
"name": "stdout",
"output_type": "stream",
"text": [
"Using Theano backend.\n"
"{'subset': 'validation', 'num_frames': 4157, 'url': 'https://www.youtube.com/watch?v=Uw_0h2UrfyY', 'duration': 139.04, 'resolution': '426x240', 'annotations': [{'segment': [19.07183775351014, 117.91560686427458], 'label': 'Ballet'}]}\n"
]
}
],
"source": [
"from work.processing.output import get_temporal_predictions_2\n",
"import numpy as np\n",
"from work.dataset.activitynet import ActivityNetDataset\n",
"import pprint\n",
"\n",
"detection_predictions_path = '../downloads/predictions/lstm_activity_detection/v1/classes/'\n",
"classification_predictions_path = '../downloads/predictions/lstm_activity_classification/v1/classes/'\n",
"\n",
"dataset = ActivityNetDataset(\n",
" videos_path='../dataset/videos.json',\n",
" labels_path='../dataset/labels.txt'\n",
")"
"with open(\"../dataset/videos.json\", \"r\") as f:\n",
" videos_info = json.load(f)\n",
"video_info = videos_info['Uw_0h2UrfyY']\n",
"print(video_info)"
]
},
{
"cell_type": "code",
"execution_count": 2,
"execution_count": 31,
"metadata": {
"collapsed": false
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 177, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]\n",
"[[ 0. 0. 0. ..., 0. 0. 0.]\n",
" [ 0. 0. 0. ..., 0. 0. 0.]\n",
" [ 0. 0. 0. ..., 0. 0. 0.]\n",
" ..., \n",
" [ 1. 0. 0. ..., 0. 0. 0.]\n",
" [ 1. 0. 0. ..., 0. 0. 0.]\n",
" [ 1. 0. 0. ..., 0. 0. 0.]]\n"
]
}
],
"source": [
"instances = generate_output(video_info, labels, length=16)\n",
"print(instances)\n",
"Y = to_categorical(instances, nb_classes=200)\n",
"print(Y[100:300])"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Old test with old code implemented."
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {
"collapsed": false,
"deletable": true,
"editable": true
},
"outputs": [
{
"name": "stdout",
Expand All @@ -57,7 +179,9 @@
"cell_type": "code",
"execution_count": 3,
"metadata": {
"collapsed": false
"collapsed": false,
"deletable": true,
"editable": true
},
"outputs": [
{
Expand Down Expand Up @@ -92,7 +216,9 @@
"cell_type": "code",
"execution_count": 4,
"metadata": {
"collapsed": false
"collapsed": false,
"deletable": true,
"editable": true
},
"outputs": [
{
Expand Down Expand Up @@ -139,7 +265,9 @@
"cell_type": "code",
"execution_count": 5,
"metadata": {
"collapsed": false
"collapsed": false,
"deletable": true,
"editable": true
},
"outputs": [
{
Expand Down Expand Up @@ -239,7 +367,9 @@
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": true
"collapsed": true,
"deletable": true,
"editable": true
},
"outputs": [],
"source": []
Expand All @@ -261,7 +391,7 @@
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.5.1"
"version": "3.6.0"
}
},
"nbformat": 4,
Expand Down
6 changes: 4 additions & 2 deletions scripts/create_stateful_dataset.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -3,14 +3,16 @@
import os
import random

import h5py
import numpy as np
from progressbar import ProgressBar

import h5py
from src.data import generate_output, import_labels, to_categorical


def create_stateful_dataset(video_features_file, videos_info, labels, output_path, batch_size, timesteps, subset=None):
def create_stateful_dataset(video_features_file, videos_info, labels,
output_path, batch_size, timesteps,
subset=None):
features_size = 4096
output_size = 201

Expand Down
8 changes: 4 additions & 4 deletions src/data.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -31,8 +31,8 @@ def to_categorical(y, nb_classes=None):
return Y

def generate_output(video_info, labels, length=16):
''' Given the info of the vide, generate a vector of classes corresponding the output for each
clip of the video which features have been extracted.
''' Given the info of the vide, generate a vector of classes corresponding
the output for each clip of the video which features have been extracted.
'''
nb_frames = video_info['num_frames']
last_first_name = nb_frames - length + 1
Expand All @@ -46,7 +46,7 @@ def generate_output(video_info, labels, length=16):
t = i / float(nb_frames) * video_info['duration']
for annotation in video_info['annotations']:
if t >= annotation['segment'][0] and t <= annotation['segment'][1]:
outputs[i] = labels.index(annotation['label'])
outputs[i] = annotation['label']
label = annotation['label']
break

Expand All @@ -59,7 +59,7 @@ def generate_output(video_info, labels, length=16):
if outs.count(label) >= length / 2:
output = labels.index(label)
else:
output = labels[0]
output = 0
instances.append(output)

return instances
Expand Down

0 comments on commit 72bcb7f

Please sign in to comment.