microsoft
/
singleshotpose
зеркало из https://github.com/microsoft/singleshotpose.git
1
0
Форкнуть 0
Код Задачи Пакеты Проекты Релизы Вики Активность

add cfg for multi-obj

This commit is contained in:
Bugra Tekin 2019-10-21 10:13:13 +02:00
Родитель 0631ec5451
Коммит 0c86c649a6
12 изменённых файлов: 207 добавлений и 226 удалений
Показать статистику Скачать Patch файл Скачать Diff файл Показать все файлы Свернуть все файлы

7
multi_obj_pose_estimation/cfg/ape_occlusion.data
Просмотреть файл

@ -3,3 +3,10 @@ mesh = ../LINEMOD/ape/ape.ply
backup = backup_multi
name = ape
diam = 0.103
gpus = 0
im_width = 640
im_height = 480
fx = 572.4114
fy = 573.5704
u0 = 325.2611
v0 = 242.0489

9
multi_obj_pose_estimation/cfg/benchvise.data
Просмотреть файл

@ -4,4 +4,11 @@ backup = backup_multi
mesh = ../LINEMOD/benchvise/benchvise.ply
tr_range = ../LINEMOD/benchvise/training_range.txt
name = benchvise
diam = 0.286908
diam = 0.286908
gpus = 0
im_width = 640
im_height = 480
fx = 572.4114
fy = 573.5704
u0 = 325.2611
v0 = 242.0489

9
multi_obj_pose_estimation/cfg/can_occlusion.data
Просмотреть файл

@ -2,4 +2,11 @@ valid = ../LINEMOD/can/test_occlusion.txt
mesh = ../LINEMOD/can/can.ply
backup = backup_multi
name = can
diam = 0.202
diam = 0.202
gpus = 0
im_width = 640
im_height = 480
fx = 572.4114
fy = 573.5704
u0 = 325.2611
v0 = 242.0489

9
multi_obj_pose_estimation/cfg/cat_occlusion.data
Просмотреть файл

@ -2,4 +2,11 @@ valid = ../LINEMOD/cat/test_occlusion.txt
mesh = ../LINEMOD/cat/cat.ply
backup = backup_multi
name = cat
diam = 0.155
diam = 0.155
gpus = 0
im_width = 640
im_height = 480
fx = 572.4114
fy = 573.5704
u0 = 325.2611
v0 = 242.0489

9
multi_obj_pose_estimation/cfg/driller_occlusion.data
Просмотреть файл

@ -2,4 +2,11 @@ valid = ../LINEMOD/driller/test_occlusion.txt
mesh = ../LINEMOD/driller/driller.ply
backup = backup_multi
name = driller
diam = 0.262
diam = 0.262
gpus = 0
im_width = 640
im_height = 480
fx = 572.4114
fy = 573.5704
u0 = 325.2611
v0 = 242.0489

9
multi_obj_pose_estimation/cfg/duck_occlusion.data
Просмотреть файл

@ -2,4 +2,11 @@ valid = ../LINEMOD/duck/test_occlusion.txt
mesh = ../LINEMOD/duck/duck.ply
backup = backup_multi
name = duck
diam = 0.109
diam = 0.109
gpus = 0
im_width = 640
im_height = 480
fx = 572.4114
fy = 573.5704
u0 = 325.2611
v0 = 242.0489

9
multi_obj_pose_estimation/cfg/eggbox_occlusion.data
Просмотреть файл

@ -2,4 +2,11 @@ valid = ../LINEMOD/eggbox/test_occlusion.txt
mesh = ../LINEMOD/eggbox/eggbox.ply
backup = backup_multi
name = eggbox
diam = 0.176364
diam = 0.176364
gpus = 0
im_width = 640
im_height = 480
fx = 572.4114
fy = 573.5704
u0 = 325.2611
v0 = 242.0489

9
multi_obj_pose_estimation/cfg/glue_occlusion.data
Просмотреть файл

@ -2,4 +2,11 @@ valid = ../LINEMOD/glue/test_occlusion.txt
mesh = ../LINEMOD/glue/glue.ply
backup = backup_multi
name = glue
diam = 0.176
diam = 0.176
gpus = 0
im_width = 640
im_height = 480
fx = 572.4114
fy = 573.5704
u0 = 325.2611
v0 = 242.0489

9
multi_obj_pose_estimation/cfg/holepuncher_occlusion.data
Просмотреть файл

@ -2,4 +2,11 @@ valid = ../LINEMOD/holepuncher/test_occlusion.txt
mesh = ../LINEMOD/holepuncher/holepuncher.ply
backup = backup_multi
name = holepuncher
diam = 0.162
diam = 0.162
gpus = 0
im_width = 640
im_height = 480
fx = 572.4114
fy = 573.5704
u0 = 325.2611
v0 = 242.0489

7
multi_obj_pose_estimation/cfg/occlusion.data
Просмотреть файл

@ -21,3 +21,10 @@ diam6 = 0.262
diam7 = 0.109
diam9 = 0.176
diam10 = 0.162
gpus = 0
im_width = 640
im_height = 480
fx = 572.4114
fy = 573.5704
u0 = 325.2611
v0 = 242.0489

11
multi_obj_pose_estimation/cfg/yolo-pose-multi.cfg
Просмотреть файл

@ -1,27 +1,24 @@
[net]
# Testing
batch=32
subdivisions=8
# Training
# batch=64
# subdivisions=8
height=416
width=416
channels=3
num_keypoints=9
momentum=0.9
decay=0.0005
angle=0
saturation = 1.5
exposure = 1.5
hue=.1
learning_rate=0.001
burn_in=1000
max_batches = 80200
policy=steps
steps=-1,100,20000,30000
# steps=-1,180,360,540
scales=0.1,10,.1,.1
conf_thresh = 0.05
max_epochs = 500
[convolutional]
batch_normalize=1
@ -235,13 +232,11 @@ activation=leaky
size=1
stride=1
pad=1
# filters=125
filters=160
activation=linear
[region]
# anchors = 1.3221, 1.73145, 3.19275, 4.00944, 5.05587, 8.09892, 9.47112, 4.84053, 11.2364, 10.0071
anchors = 1.4820, 2.2412, 2.0501, 3.1265, 2.3946, 4.6891, 3.1018, 3.9910, 3.4879, 5.8851
bias_match=1
classes=13

336
multi_obj_pose_estimation/valid_multi.ipynb
Просмотреть файл

@ -2,203 +2,137 @@
"cells": [
{
"cell_type": "code",
"execution_count": 1,
"execution_count": 4,
"metadata": {},
"outputs": [],
"source": [
"%matplotlib inline\n",
"import os\n",
"os.sys.path.append('..')\n",
"os.environ[\"CUDA_VISIBLE_DEVICES\"]=\"1\"\n",
"import torch\n",
"from torch.autograd import Variable\n",
"from torchvision import datasets, transforms\n",
"from scipy.misc import imsave\n",
"import scipy.io\n",
"import warnings\n",
"import sys\n",
"warnings.filterwarnings(\"ignore\")\n",
"import matplotlib.pyplot as plt\n",
"import scipy.misc\n",
"import warnings\n",
"import sys\n",
"import argparse\n",
"warnings.filterwarnings(\"ignore\")\n",
"from torch.autograd import Variable\n",
"from torchvision import datasets, transforms\n",
"\n",
"from darknet_multi import Darknet\n",
"from utils import *\n",
"import dataset_multi\n",
"from MeshPly import MeshPly"
"from darknet_multi import Darknet\n",
"from utils_multi import *\n",
"from cfg import parse_cfg\n",
"from MeshPly import MeshPly\n",
"\n",
"import matplotlib.pyplot as plt\n",
"from scipy.misc import imsave\n",
"import scipy.io\n",
"import scipy.misc"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "code",
"execution_count": 2,
"execution_count": 5,
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"2018-05-06 14:09:50 Testing ape...\n",
"2018-05-06 14:10:15 Acc using 5 px 2D Projection = 7.01%\n",
"2018-05-06 14:10:15 Acc using 10 px 2D Projection = 40.43%\n",
"2018-05-06 14:10:15 Acc using 15 px 2D Projection = 59.83%\n",
"2018-05-06 14:10:15 Acc using 20 px 2D Projection = 68.55%\n",
"2018-05-06 14:10:15 Acc using 25 px 2D Projection = 72.05%\n",
"2018-05-06 14:10:15 Acc using 30 px 2D Projection = 73.68%\n",
"2018-05-06 14:10:15 Acc using 35 px 2D Projection = 74.53%\n",
"2018-05-06 14:10:15 Acc using 40 px 2D Projection = 75.13%\n",
"2018-05-06 14:10:15 Acc using 45 px 2D Projection = 75.73%\n",
"2018-05-06 14:10:15 Acc using 50 px 2D Projection = 76.50%\n",
"2018-05-06 14:10:18 Testing can...\n",
"2018-05-06 14:10:47 Acc using 5 px 2D Projection = 11.18%\n",
"2018-05-06 14:10:47 Acc using 10 px 2D Projection = 57.83%\n",
"2018-05-06 14:10:47 Acc using 15 px 2D Projection = 79.95%\n",
"2018-05-06 14:10:47 Acc using 20 px 2D Projection = 85.75%\n",
"2018-05-06 14:10:47 Acc using 25 px 2D Projection = 88.73%\n",
"2018-05-06 14:10:47 Acc using 30 px 2D Projection = 90.39%\n",
"2018-05-06 14:10:47 Acc using 35 px 2D Projection = 91.80%\n",
"2018-05-06 14:10:47 Acc using 40 px 2D Projection = 93.21%\n",
"2018-05-06 14:10:47 Acc using 45 px 2D Projection = 93.62%\n",
"2018-05-06 14:10:47 Acc using 50 px 2D Projection = 93.79%\n",
"2018-05-06 14:10:50 Testing cat...\n",
"2018-05-06 14:11:16 Acc using 5 px 2D Projection = 3.62%\n",
"2018-05-06 14:11:16 Acc using 10 px 2D Projection = 23.25%\n",
"2018-05-06 14:11:16 Acc using 15 px 2D Projection = 39.51%\n",
"2018-05-06 14:11:16 Acc using 20 px 2D Projection = 49.45%\n",
"2018-05-06 14:11:16 Acc using 25 px 2D Projection = 54.76%\n",
"2018-05-06 14:11:16 Acc using 30 px 2D Projection = 57.96%\n",
"2018-05-06 14:11:16 Acc using 35 px 2D Projection = 59.56%\n",
"2018-05-06 14:11:16 Acc using 40 px 2D Projection = 60.99%\n",
"2018-05-06 14:11:16 Acc using 45 px 2D Projection = 62.51%\n",
"2018-05-06 14:11:16 Acc using 50 px 2D Projection = 63.27%\n",
"2018-05-06 14:11:19 Testing duck...\n",
"2018-05-06 14:11:42 Acc using 5 px 2D Projection = 5.07%\n",
"2018-05-06 14:11:42 Acc using 10 px 2D Projection = 18.20%\n",
"2018-05-06 14:11:42 Acc using 15 px 2D Projection = 30.88%\n",
"2018-05-06 14:11:42 Acc using 20 px 2D Projection = 55.12%\n",
"2018-05-06 14:11:42 Acc using 25 px 2D Projection = 75.15%\n",
"2018-05-06 14:11:42 Acc using 30 px 2D Projection = 81.45%\n",
"2018-05-06 14:11:42 Acc using 35 px 2D Projection = 83.20%\n",
"2018-05-06 14:11:42 Acc using 40 px 2D Projection = 83.64%\n",
"2018-05-06 14:11:42 Acc using 45 px 2D Projection = 83.90%\n",
"2018-05-06 14:11:42 Acc using 50 px 2D Projection = 84.16%\n",
"2018-05-06 14:11:45 Testing driller...\n",
"2018-05-06 14:12:10 Acc using 5 px 2D Projection = 1.40%\n",
"2018-05-06 14:12:10 Acc using 10 px 2D Projection = 17.38%\n",
"2018-05-06 14:12:10 Acc using 15 px 2D Projection = 39.87%\n",
"2018-05-06 14:12:10 Acc using 20 px 2D Projection = 62.93%\n",
"2018-05-06 14:12:10 Acc using 25 px 2D Projection = 80.64%\n",
"2018-05-06 14:12:10 Acc using 30 px 2D Projection = 89.87%\n",
"2018-05-06 14:12:10 Acc using 35 px 2D Projection = 94.89%\n",
"2018-05-06 14:12:10 Acc using 40 px 2D Projection = 95.88%\n",
"2018-05-06 14:12:10 Acc using 45 px 2D Projection = 96.54%\n",
"2018-05-06 14:12:10 Acc using 50 px 2D Projection = 96.87%\n",
"2018-05-06 14:12:13 Testing glue...\n",
"2018-05-06 14:12:31 Acc using 5 px 2D Projection = 6.53%\n",
"2018-05-06 14:12:31 Acc using 10 px 2D Projection = 26.91%\n",
"2018-05-06 14:12:31 Acc using 15 px 2D Projection = 39.65%\n",
"2018-05-06 14:12:31 Acc using 20 px 2D Projection = 46.18%\n",
"2018-05-06 14:12:31 Acc using 25 px 2D Projection = 49.50%\n",
"2018-05-06 14:12:31 Acc using 30 px 2D Projection = 51.83%\n",
"2018-05-06 14:12:31 Acc using 35 px 2D Projection = 53.05%\n",
"2018-05-06 14:12:31 Acc using 40 px 2D Projection = 53.16%\n",
"2018-05-06 14:12:31 Acc using 45 px 2D Projection = 53.93%\n",
"2018-05-06 14:12:31 Acc using 50 px 2D Projection = 54.71%\n",
"2018-05-06 14:12:45 Testing holepuncher...\n",
"2018-05-06 14:19:31 Acc using 5 px 2D Projection = 8.26%\n",
"2018-05-06 14:19:31 Acc using 10 px 2D Projection = 39.50%\n",
"2018-05-06 14:19:31 Acc using 15 px 2D Projection = 53.31%\n",
"2018-05-06 14:19:31 Acc using 20 px 2D Projection = 62.56%\n",
"2018-05-06 14:19:31 Acc using 25 px 2D Projection = 68.02%\n",
"2018-05-06 14:19:31 Acc using 30 px 2D Projection = 74.71%\n",
"2018-05-06 14:19:31 Acc using 35 px 2D Projection = 80.74%\n",
"2018-05-06 14:19:31 Acc using 40 px 2D Projection = 85.62%\n",
"2018-05-06 14:19:31 Acc using 45 px 2D Projection = 89.59%\n",
"2018-05-06 14:19:31 Acc using 50 px 2D Projection = 91.49%\n"
"2019-10-18 17:00:04 Testing ape...\n",
"2019-10-18 17:01:38 Acc using 5 px 2D Projection = 6.07%\n",
"2019-10-18 17:01:38 Acc using 10 px 2D Projection = 39.32%\n",
"2019-10-18 17:01:38 Acc using 15 px 2D Projection = 59.83%\n",
"2019-10-18 17:01:38 Acc using 20 px 2D Projection = 68.29%\n",
"2019-10-18 17:01:38 Acc using 25 px 2D Projection = 72.74%\n",
"2019-10-18 17:01:38 Acc using 30 px 2D Projection = 74.96%\n",
"2019-10-18 17:01:38 Acc using 35 px 2D Projection = 75.64%\n",
"2019-10-18 17:01:38 Acc using 40 px 2D Projection = 76.32%\n",
"2019-10-18 17:01:38 Acc using 45 px 2D Projection = 76.67%\n",
"2019-10-18 17:01:38 Acc using 50 px 2D Projection = 78.03%\n",
"2019-10-18 17:01:39 Testing can...\n"
]
},
{
"ename": "KeyboardInterrupt",
"evalue": "",
"output_type": "error",
"traceback": [
"\u001b[1;31m---------------------------------------------------------------------------\u001b[0m",
"\u001b[1;31mKeyboardInterrupt\u001b[0m Traceback (most recent call last)",
"\u001b[1;32m<ipython-input-5-27dc0bb0bf4c>\u001b[0m in \u001b[0;36m<module>\u001b[1;34m()\u001b[0m\n\u001b[0;32m 146\u001b[0m \u001b[0mvalid\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mdatacfg\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mmodelcfg\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mweightfile\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m 147\u001b[0m \u001b[0mdatacfg\u001b[0m \u001b[1;33m=\u001b[0m \u001b[1;34m'cfg/can_occlusion.data'\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[1;32m--> 148\u001b[1;33m \u001b[0mvalid\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mdatacfg\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mmodelcfg\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mweightfile\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0m\u001b[0;32m 149\u001b[0m \u001b[0mdatacfg\u001b[0m \u001b[1;33m=\u001b[0m \u001b[1;34m'cfg/cat_occlusion.data'\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m 150\u001b[0m \u001b[0mvalid\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mdatacfg\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mmodelcfg\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mweightfile\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n",
"\u001b[1;32m<ipython-input-5-27dc0bb0bf4c>\u001b[0m in \u001b[0;36mvalid\u001b[1;34m(datacfg, cfgfile, weightfile)\u001b[0m\n\u001b[0;32m 74\u001b[0m \u001b[1;31m# Using confidence threshold, eliminate low-confidence predictions\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m 75\u001b[0m \u001b[0mtrgt\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0mtarget\u001b[0m\u001b[1;33m[\u001b[0m\u001b[1;36m0\u001b[0m\u001b[1;33m]\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mview\u001b[0m\u001b[1;33m(\u001b[0m\u001b[1;33m-\u001b[0m\u001b[1;36m1\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mnum_labels\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[1;32m---> 76\u001b[1;33m \u001b[0mall_boxes\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0mget_multi_region_boxes\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0moutput\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mconf_thresh\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mnum_classes\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mnum_keypoints\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0manchors\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mnum_anchors\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0mint\u001b[0m\u001b[1;33m(\u001b[0m\u001b[0mtrgt\u001b[0m\u001b[1;33m[\u001b[0m\u001b[1;36m0\u001b[0m\u001b[1;33m]\u001b[0m\u001b[1;33m[\u001b[0m\u001b[1;36m0\u001b[0m\u001b[1;33m]\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m,\u001b[0m \u001b[0monly_objectness\u001b[0m\u001b[1;33m=\u001b[0m\u001b[1;36m0\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0m\u001b[0;32m 77\u001b[0m \u001b[0mt4\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0mtime\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mtime\u001b[0m\u001b[1;33m(\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m 78\u001b[0m \u001b[1;33m\u001b[0m\u001b[0m\n",
"\u001b[1;32m~\\Documents\\Code\\singleshot6Dpose\\multi_obj_pose_estimation\\utils_multi.py\u001b[0m in \u001b[0;36mget_multi_region_boxes\u001b[1;34m(output, conf_thresh, num_classes, num_keypoints, anchors, num_anchors, correspondingclass, only_objectness, validation)\u001b[0m\n\u001b[0;32m 330\u001b[0m \u001b[0mmax_ind\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0mind\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m 331\u001b[0m \u001b[1;33m\u001b[0m\u001b[0m\n\u001b[1;32m--> 332\u001b[1;33m \u001b[1;32mif\u001b[0m \u001b[0mconf\u001b[0m \u001b[1;33m>\u001b[0m \u001b[0mconf_thresh\u001b[0m\u001b[1;33m:\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0m\u001b[0;32m 333\u001b[0m \u001b[0mbcx\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0mlist\u001b[0m\u001b[1;33m(\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m 334\u001b[0m \u001b[0mbcy\u001b[0m \u001b[1;33m=\u001b[0m \u001b[0mlist\u001b[0m\u001b[1;33m(\u001b[0m\u001b[1;33m)\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n",
"\u001b[1;31mKeyboardInterrupt\u001b[0m: "
]
}
],
"source": [
"def valid(datacfg, cfgfile, weightfile, conf_th):\n",
"def valid(datacfg, cfgfile, weightfile):\n",
" def truths_length(truths):\n",
" for i in range(50):\n",
" if truths[i][1] == 0:\n",
" return i\n",
"\n",
" # Parse configuration files\n",
" options = read_data_cfg(datacfg)\n",
" valid_images = options['valid']\n",
" meshname = options['mesh']\n",
" backupdir = options['backup']\n",
" name = options['name']\n",
" prefix = 'results'\n",
" # Read object model information, get 3D bounding box corners\n",
" mesh = MeshPly(meshname)\n",
" vertices = np.c_[np.array(mesh.vertices), np.ones((len(mesh.vertices), 1))].transpose()\n",
" corners3D = get_3D_corners(vertices)\n",
" # Read intrinsic camera parameters\n",
" internal_calibration = get_camera_intrinsic()\n",
" # Parse data configuration files\n",
" data_options = read_data_cfg(datacfg)\n",
" valid_images = data_options['valid']\n",
" meshname = data_options['mesh']\n",
" name = data_options['name']\n",
" im_width = int(data_options['im_width'])\n",
" im_height = int(data_options['im_height']) \n",
" fx = float(data_options['fx'])\n",
" fy = float(data_options['fy'])\n",
" u0 = float(data_options['u0'])\n",
" v0 = float(data_options['v0'])\n",
" \n",
" # Parse net configuration file\n",
" net_options = parse_cfg(cfgfile)[0]\n",
" loss_options = parse_cfg(cfgfile)[-1]\n",
" conf_thresh = float(net_options['conf_thresh'])\n",
" num_keypoints = int(net_options['num_keypoints'])\n",
" num_classes = int(loss_options['classes'])\n",
" num_anchors = int(loss_options['num'])\n",
" anchors = [float(anchor) for anchor in loss_options['anchors'].split(',')]\n",
"\n",
" # Get validation file names\n",
" with open(valid_images) as fp:\n",
" # Read object model information, get 3D bounding box corners, get intrinsics\n",
" mesh = MeshPly(meshname)\n",
" vertices = np.c_[np.array(mesh.vertices), np.ones((len(mesh.vertices), 1))].transpose()\n",
" corners3D = get_3D_corners(vertices)\n",
" diam = float(data_options['diam'])\n",
" intrinsic_calibration = get_camera_intrinsic(u0, v0, fx, fy) # camera params\n",
"\n",
" # Network I/O params\n",
" num_labels = 2*num_keypoints+3 # +2 for width, height, +1 for object class\n",
" errs_2d = [] # to save\n",
" with open(valid_images) as fp: # validation file names\n",
" tmp_files = fp.readlines()\n",
" valid_files = [item.rstrip() for item in tmp_files]\n",
" \n",
"\n",
" # Compute-related Parameters\n",
" use_cuda = True # whether to use cuda or no\n",
" kwargs = {'num_workers': 4, 'pin_memory': True} # number of workers etc.\n",
"\n",
" # Specicy model, load pretrained weights, pass to GPU and set the module in evaluation mode\n",
" model = Darknet(cfgfile)\n",
" model.load_weights(weightfile)\n",
" model.cuda()\n",
" model.eval()\n",
"\n",
" # Get the parser for the test dataset\n",
" valid_dataset = dataset_multi.listDataset(valid_images, shape=(model.width, model.height),\n",
" shuffle=False,\n",
" objclass=name,\n",
" transform=transforms.Compose([\n",
" transforms.ToTensor(),\n",
" ]))\n",
" valid_batchsize = 1\n",
"\n",
" # Specify the number of workers for multiple processing, get the dataloader for the test dataset\n",
" kwargs = {'num_workers': 4, 'pin_memory': True}\n",
" test_loader = torch.utils.data.DataLoader(\n",
" valid_dataset, batch_size=valid_batchsize, shuffle=False, **kwargs) \n",
"\n",
" # Parameters\n",
" visualize = False\n",
" use_cuda = True\n",
" num_classes = 13\n",
" anchors = [1.4820, 2.2412, 2.0501, 3.1265, 2.3946, 4.6891, 3.1018, 3.9910, 3.4879, 5.8851]\n",
" num_anchors = 5\n",
" eps = 1e-5\n",
" conf_thresh = conf_th\n",
" iou_thresh = 0.5\n",
"\n",
" # Parameters to save\n",
" errs_2d = []\n",
" edges = [[1, 2], [1, 3], [1, 5], [2, 4], [2, 6], [3, 4], [3, 7], [4, 8], [5, 6], [5, 7], [6, 8], [7, 8]]\n",
" edges_corners = [[0, 1], [0, 2], [0, 4], [1, 3], [1, 5], [2, 3], [2, 6], [3, 7], [4, 5], [4, 6], [5, 7], [6, 7]]\n",
" # Get the dataloader for the test dataset\n",
" valid_dataset = dataset_multi.listDataset(valid_images, shape=(model.width, model.height), shuffle=False, objclass=name, transform=transforms.Compose([transforms.ToTensor(),]))\n",
" test_loader = torch.utils.data.DataLoader(valid_dataset, batch_size=1, shuffle=False, **kwargs) \n",
"\n",
" # Iterate through test batches (Batch size for test data is 1)\n",
" count = 0\n",
" logging('Testing {}...'.format(name))\n",
" for batch_idx, (data, target) in enumerate(test_loader):\n",
" \n",
" # Images\n",
" img = data[0, :, :, :]\n",
" img = img.numpy().squeeze()\n",
" img = np.transpose(img, (1, 2, 0))\n",
" \n",
" t1 = time.time()\n",
" # Pass data to GPU\n",
" if use_cuda:\n",
" data = data.cuda()\n",
" target = target.cuda()\n",
" # target = target.cuda()\n",
" \n",
" # Wrap tensors in Variable class, set volatile=True for inference mode and to use minimal memory during inference\n",
" data = Variable(data, volatile=True)\n",
@ -209,8 +143,8 @@
" t3 = time.time()\n",
" \n",
" # Using confidence threshold, eliminate low-confidence predictions\n",
" trgt = target[0].view(-1, 21)\n",
" all_boxes = get_corresponding_region_boxes(output, conf_thresh, num_classes, anchors, num_anchors, int(trgt[0][0]), only_objectness=0) \n",
" trgt = target[0].view(-1, num_labels)\n",
" all_boxes = get_multi_region_boxes(output, conf_thresh, num_classes, num_keypoints, anchors, num_anchors, int(trgt[0][0]), only_objectness=0) \n",
" t4 = time.time()\n",
" \n",
" # Iterate through all images in the batch\n",
@ -220,95 +154,77 @@
" boxes = all_boxes[i]\n",
" \n",
" # For each image, get all the targets (for multiple object pose estimation, there might be more than 1 target per image)\n",
" truths = target[i].view(-1, 21)\n",
" truths = target[i].view(-1, num_labels)\n",
" \n",
" # Get how many object are present in the scene\n",
" num_gts = truths_length(truths)\n",
"\n",
" # Iterate through each ground-truth object\n",
" for k in range(num_gts):\n",
" box_gt = [truths[k][1], truths[k][2], truths[k][3], truths[k][4], truths[k][5], truths[k][6], \n",
" truths[k][7], truths[k][8], truths[k][9], truths[k][10], truths[k][11], truths[k][12], \n",
" truths[k][13], truths[k][14], truths[k][15], truths[k][16], truths[k][17], truths[k][18], 1.0, 1.0, truths[k][0]]\n",
" best_conf_est = -1\n",
" box_gt = list()\n",
" for j in range(1, num_labels):\n",
" box_gt.append(truths[k][j])\n",
" box_gt.extend([1.0, 1.0])\n",
" box_gt.append(truths[k][0])\n",
" \n",
"\n",
" # If the prediction has the highest confidence, choose it as our prediction\n",
" best_conf_est = -sys.maxsize\n",
" for j in range(len(boxes)):\n",
" if (boxes[j][18] > best_conf_est) and (boxes[j][20] == int(truths[k][0])):\n",
" best_conf_est = boxes[j][18]\n",
" if (boxes[j][2*num_keypoints] > best_conf_est) and (boxes[j][2*num_keypoints+2] == int(truths[k][0])):\n",
" best_conf_est = boxes[j][2*num_keypoints]\n",
" box_pr = boxes[j]\n",
" bb2d_gt = get_2d_bb(box_gt[:18], output.size(3))\n",
" bb2d_pr = get_2d_bb(box_pr[:18], output.size(3))\n",
" iou = bbox_iou(bb2d_gt, bb2d_pr)\n",
" match = corner_confidence9(box_gt[:18], torch.FloatTensor(boxes[j][:18]))\n",
" match = corner_confidence(box_gt[:2*num_keypoints], torch.FloatTensor(boxes[j][:2*num_keypoints]))\n",
" \n",
" # Denormalize the corner predictions \n",
" corners2D_gt = np.array(np.reshape(box_gt[:18], [9, 2]), dtype='float32')\n",
" corners2D_pr = np.array(np.reshape(box_pr[:18], [9, 2]), dtype='float32')\n",
" corners2D_gt[:, 0] = corners2D_gt[:, 0] * 640\n",
" corners2D_gt[:, 1] = corners2D_gt[:, 1] * 480 \n",
" corners2D_pr[:, 0] = corners2D_pr[:, 0] * 640\n",
" corners2D_pr[:, 1] = corners2D_pr[:, 1] * 480\n",
" corners2D_gt_corrected = fix_corner_order(corners2D_gt) # Fix the order of the corners in OCCLUSION\n",
" corners2D_gt = np.array(np.reshape(box_gt[:2*num_keypoints], [-1, 2]), dtype='float32')\n",
" corners2D_pr = np.array(np.reshape(box_pr[:2*num_keypoints], [-1, 2]), dtype='float32')\n",
" corners2D_gt[:, 0] = corners2D_gt[:, 0] * im_width\n",
" corners2D_gt[:, 1] = corners2D_gt[:, 1] * im_height \n",
" corners2D_pr[:, 0] = corners2D_pr[:, 0] * im_width\n",
" corners2D_pr[:, 1] = corners2D_pr[:, 1] * im_height\n",
" corners2D_gt_corrected = fix_corner_order(corners2D_gt) # Fix the order of corners\n",
" \n",
" # Compute [R|t] by pnp\n",
" objpoints3D = np.array(np.transpose(np.concatenate((np.zeros((3, 1)), corners3D[:3, :]), axis=1)), dtype='float32')\n",
" K = np.array(internal_calibration, dtype='float32')\n",
" K = np.array(intrinsic_calibration, dtype='float32')\n",
" R_gt, t_gt = pnp(objpoints3D, corners2D_gt_corrected, K)\n",
" R_pr, t_pr = pnp(objpoints3D, corners2D_pr, K)\n",
" \n",
" # Compute pixel error\n",
" Rt_gt = np.concatenate((R_gt, t_gt), axis=1)\n",
" Rt_pr = np.concatenate((R_pr, t_pr), axis=1)\n",
" proj_2d_gt = compute_projection(vertices, Rt_gt, internal_calibration) \n",
" proj_2d_pred = compute_projection(vertices, Rt_pr, internal_calibration) \n",
" proj_corners_gt = np.transpose(compute_projection(corners3D, Rt_gt, internal_calibration)) \n",
" proj_corners_pr = np.transpose(compute_projection(corners3D, Rt_pr, internal_calibration)) \n",
" proj_2d_gt = compute_projection(vertices, Rt_gt, intrinsic_calibration) \n",
" proj_2d_pred = compute_projection(vertices, Rt_pr, intrinsic_calibration) \n",
" proj_corners_gt = np.transpose(compute_projection(corners3D, Rt_gt, intrinsic_calibration)) \n",
" proj_corners_pr = np.transpose(compute_projection(corners3D, Rt_pr, intrinsic_calibration)) \n",
" norm = np.linalg.norm(proj_2d_gt - proj_2d_pred, axis=0)\n",
" pixel_dist = np.mean(norm)\n",
" errs_2d.append(pixel_dist)\n",
"\n",
" \n",
" if visualize:\n",
" # Visualize\n",
" plt.xlim((0, 640))\n",
" plt.ylim((0, 480))\n",
" plt.imshow(scipy.misc.imresize(img, (480, 640)))\n",
" # Projections\n",
" for edge in edges_corners:\n",
" plt.plot(proj_corners_gt[edge, 0], proj_corners_gt[edge, 1], color='g', linewidth=3.0)\n",
" plt.plot(proj_corners_pr[edge, 0], proj_corners_pr[edge, 1], color='b', linewidth=3.0)\n",
" plt.gca().invert_yaxis()\n",
" plt.show()\n",
"\n",
" t5 = time.time()\n",
"\n",
" # Compute 2D projection score\n",
" eps = 1e-5\n",
" for px_threshold in [5, 10, 15, 20, 25, 30, 35, 40, 45, 50]:\n",
" acc = len(np.where(np.array(errs_2d) <= px_threshold)[0]) * 100. / (len(errs_2d)+eps)\n",
" # Print test statistics\n",
" logging(' Acc using {} px 2D Projection = {:.2f}%'.format(px_threshold, acc))\n",
"\n",
"conf_th = 0.05\n",
"cfgfile = 'cfg/yolo-pose-multi.cfg'\n",
"weightfile = 'backup_multi/model_backup2.weights'\n",
"modelcfg = 'cfg/yolo-pose-multi.cfg'\n",
"datacfg = 'cfg/ape_occlusion.data'\n",
"valid(datacfg, cfgfile, weightfile, conf_th)\n",
"datacfg = 'cfg/can_occlusion.data'\n",
"valid(datacfg, cfgfile, weightfile, conf_th)\n",
"datacfg = 'cfg/cat_occlusion.data'\n",
"valid(datacfg, cfgfile, weightfile, conf_th)\n",
"datacfg = 'cfg/duck_occlusion.data'\n",
"valid(datacfg, cfgfile, weightfile, conf_th)\n",
"datacfg = 'cfg/driller_occlusion.data'\n",
"valid(datacfg, cfgfile, weightfile, conf_th)\n",
"datacfg = 'cfg/glue_occlusion.data'\n",
"valid(datacfg, cfgfile, weightfile, conf_th)\n",
"datacfg = 'cfg/holepuncher_occlusion.data'\n",
"valid(datacfg, cfgfile, weightfile, conf_th)\n",
"weightfile = 'backup_multi/model_backup.weights'\n",
"\n",
" "
"valid(datacfg, modelcfg, weightfile)\n",
"datacfg = 'cfg/can_occlusion.data'\n",
"valid(datacfg, modelcfg, weightfile)\n",
"datacfg = 'cfg/cat_occlusion.data'\n",
"valid(datacfg, modelcfg, weightfile)\n",
"datacfg = 'cfg/duck_occlusion.data'\n",
"valid(datacfg, modelcfg, weightfile)\n",
"datacfg = 'cfg/glue_occlusion.data'\n",
"valid(datacfg, modelcfg, weightfile)\n",
"datacfg = 'cfg/holepuncher_occlusion.data'\n",
"valid(datacfg, modelcfg, weightfile)"
]
},
{
@ -321,21 +237,21 @@
],
"metadata": {
"kernelspec": {
"display_name": "Python 2",
"display_name": "Python 3",
"language": "python",
"name": "python2"
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 2
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython2",
"version": "2.7.12"
"pygments_lexer": "ipython3",
"version": "3.6.6"
}
},
"nbformat": 4,