Movatterモバイル変換

[0]ホーム
URL:

Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

 Sign up
Appearance settings

Commitd66cc2b

Browse files
committed
MS COCO dataset support
1 parent4520c19 commitd66cc2b

File tree

90 files changed

+4596
-205
lines changed

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.

90 files changed

+4596
-205
lines changed

‎README.md‎

Lines changed: 18 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -55,6 +55,8 @@ If you find Faster R-CNN useful in your research, please consider citing:
5555
```make
5656
# In your Makefile.config, make sure to have this line uncommented
5757
WITH_PYTHON_LAYER := 1
58+
# Unrelatedly, it's also recommended that you use CUDNN
59+
USE_CUDNN := 1
5860
```
5961

6062
You can download my[Makefile.config](http://www.cs.berkeley.edu/~rbg/fast-rcnn-data/Makefile.config) for reference.
@@ -115,8 +117,6 @@ If you find Faster R-CNN useful in your research, please consider citing:
115117

116118
*After successfully completing [basic installation](#installation-sufficient-for-the-demo)*, you'll be ready to run the demo.
117119

118-
**Python**
119-
120120
To run the demo
121121
```Shell
122122
cd$FRCN_ROOT
@@ -158,7 +158,8 @@ The demo performs detection using a VGG16 network trained for detection on PASCA
158158
```
159159
Using symlinks is a good idea because you will likely want to share the same PASCAL dataset installation between multiple projects.
160160
5. [Optional] follow similar steps to get PASCAL VOC 2010 and 2012
161-
6. Follow the next sections to download pre-trained ImageNet models
161+
6. [Optional] If you want to use COCO, please see some notes under`data/README.md`
162+
7. Follow the next sections to download pre-trained ImageNet models
162163

163164
### Download pre-trained ImageNet models
164165

@@ -200,3 +201,17 @@ cd $FRCN_ROOT
200201
```
201202

202203
This method trains the RPN module jointly with the Fast R-CNN network, rather than alternating between training the two. It resultsin faster (~ 1.5x speedup) trainingtimes and similar detection accuracy. See these [slides](https://www.dropbox.com/s/xtr4yd4i5e0vw8g/iccv15_tutorial_training_rbg.pdf?dl=0)for more details.
204+
205+
Artifacts generated by the scriptsin`tools` are writtenin this directory.
206+
207+
Trained Fast R-CNN networks are saved under:
208+
209+
```
210+
output/<experiment directory>/<dataset name>/
211+
```
212+
213+
Test outputs are saved under:
214+
215+
```
216+
output/<experiment directory>/<dataset name>/<network snapshot name>/
217+
```

‎data/README.md‎

Lines changed: 14 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -30,6 +30,20 @@ ln -s /your/path/to/VOC2007/VOCdevkit VOCdevkit2007
3030
ln -s /your/path/to/VOC2012/VOCdevkit VOCdevkit2012
3131
```
3232

33+
Install the MS COCO dataset at /path/to/coco
34+
35+
```
36+
ln -s /path/to/coco coco
37+
```
38+
39+
For COCO with Fast R-CNN, place object proposals under`coco_proposals` (inside
40+
the`data` directory). You can obtain proposals on COCO from Jan Hosang at
41+
https://www.mpi-inf.mpg.de/departments/computer-vision-and-multimodal-computing/research/object-recognition-and-scene-understanding/how-good-are-detection-proposals-really/.
42+
For COCO, using MCG is recommended over selective search. MCG boxes can be downloaded
43+
fromhttp://www.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/mcg/.
44+
Use the tool`lib/datasets/tools/mcg_munge.py` to convert the downloaded MCG data
45+
into the same file layout as those from Jan Hosang.
46+
3347
Since you'll likely be experimenting with multiple installs of Fast/er R-CNN in
3448
parallel, you'll probably want to keep all of this data in a shared place and
3549
use symlinks. On my system I create the following symlinks inside`data`:
Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1,3 +1,5 @@
11
EXP_DIR:faster_rcnn_alt_opt
2+
TRAIN:
3+
BG_THRESH_LO:0.0
24
TEST:
35
HAS_RPN:True

‎experiments/cfgs/faster_rcnn_end2end.yml‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -6,5 +6,6 @@ TRAIN:
66
RPN_POSITIVE_OVERLAP:0.7
77
RPN_BATCHSIZE:256
88
PROPOSAL_METHOD:gt
9+
BG_THRESH_LO:0.0
910
TEST:
1011
HAS_RPN:True

‎experiments/scripts/fast_rcnn.sh‎

Lines changed: 31 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -1,8 +1,10 @@
11
#!/bin/bash
22
# Usage:
3-
# ./experiments/scripts/default.sh GPU NET [options args to {train,test}_net.py]
3+
# ./experiments/scripts/fast_rcnn.sh GPU NET DATASET [options args to {train,test}_net.py]
4+
# DATASET is either pascal_voc or coco.
5+
#
46
# Example:
5-
# ./experiments/scripts/default.sh 0CaffeNet \
7+
# ./experiments/scripts/fast_rcnn.sh 0VGG_CNN_M_1024 pascal_voc \
68
# --set EXP_DIR foobar RNG_SEED 42 TRAIN.SCALES "[400, 500, 600, 700]"
79

810
set -x
@@ -13,28 +15,49 @@ export PYTHONUNBUFFERED="True"
1315
GPU_ID=$1
1416
NET=$2
1517
NET_lc=${NET,,}
18+
DATASET=$3
1619

1720
array=($@ )
1821
len=${#array[@]}
19-
EXTRA_ARGS=${array[@]:2:$len}
22+
EXTRA_ARGS=${array[@]:3:$len}
2023
EXTRA_ARGS_SLUG=${EXTRA_ARGS///_}
2124

22-
LOG="experiments/logs/default_${NET}_${EXTRA_ARGS_SLUG}.txt.`date +'%Y-%m-%d_%H-%M-%S'`"
25+
case$DATASETin
26+
pascal_voc)
27+
TRAIN_IMDB="voc_2007_trainval"
28+
TEST_IMDB="voc_2007_test"
29+
PT_DIR="pascal_voc"
30+
ITERS=40000
31+
;;
32+
coco)
33+
TRAIN_IMDB="coco_2014_train"
34+
TEST_IMDB="coco_2014_minival"
35+
PT_DIR="coco"
36+
ITERS=280000
37+
;;
38+
*)
39+
echo"No dataset given"
40+
exit
41+
;;
42+
esac
43+
44+
LOG="experiments/logs/fast_rcnn_${NET}_${EXTRA_ARGS_SLUG}.txt.`date +'%Y-%m-%d_%H-%M-%S'`"
2345
exec&>>(tee -a"$LOG")
2446
echo Logging output to"$LOG"
2547

2648
time ./tools/train_net.py --gpu${GPU_ID} \
27-
--solver models/${NET}/fast_rcnn/solver.prototxt \
49+
--solver models/${PT_DIR}/${NET}/fast_rcnn/solver.prototxt \
2850
--weights data/imagenet_models/${NET}.v2.caffemodel \
29-
--imdb voc_2007_trainval \
51+
--imdb${TRAIN_IMDB} \
52+
--iters${ITERS} \
3053
${EXTRA_ARGS}
3154

3255
set +x
3356
NET_FINAL=`grep -B 1"done solving"${LOG}| grep"Wrote snapshot"| awk'{print $4}'`
3457
set -x
3558

3659
time ./tools/test_net.py --gpu${GPU_ID} \
37-
--def models/${NET}/fast_rcnn/test.prototxt \
60+
--def models/${PT_DIR}/${NET}/fast_rcnn/test.prototxt \
3861
--net${NET_FINAL} \
39-
--imdbvoc_2007_test \
62+
--imdb${TEST_IMDB} \
4063
${EXTRA_ARGS}
Lines changed: 27 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -1,9 +1,11 @@
11
#!/bin/bash
22
# Usage:
3-
# ./experiments/scripts/default_faster_rcnn_alt_opt.sh GPU NET [--set ...]
3+
# ./experiments/scripts/faster_rcnn_alt_opt.sh GPU NET DATASET [options args to {train,test}_net.py]
4+
# DATASET is only pascal_voc for now
5+
#
46
# Example:
5-
# ./experiments/scripts/default_faster_rcnn_alt_opt.sh 0ZF \
6-
# --set EXP_DIR foobar RNG_SEED 42 TRAIN.SCALES "[400,500,600,700]"
7+
# ./experiments/scripts/faster_rcnn_alt_opt.sh 0VGG_CNN_M_1024 pascal_voc \
8+
# --set EXP_DIR foobar RNG_SEED 42 TRAIN.SCALES "[400,500,600,700]"
79

810
set -x
911
set -e
@@ -13,20 +15,38 @@ export PYTHONUNBUFFERED="True"
1315
GPU_ID=$1
1416
NET=$2
1517
NET_lc=${NET,,}
18+
DATASET=$3
1619

1720
array=($@ )
1821
len=${#array[@]}
19-
EXTRA_ARGS=${array[@]:2:$len}
22+
EXTRA_ARGS=${array[@]:3:$len}
2023
EXTRA_ARGS_SLUG=${EXTRA_ARGS///_}
2124

25+
case$DATASETin
26+
pascal_voc)
27+
TRAIN_IMDB="voc_2007_trainval"
28+
TEST_IMDB="voc_2007_test"
29+
PT_DIR="pascal_voc"
30+
ITERS=40000
31+
;;
32+
coco)
33+
echo"Not implemented: use experiments/scripts/faster_rcnn_end2end.sh for coco"
34+
exit
35+
;;
36+
*)
37+
echo"No dataset given"
38+
exit
39+
;;
40+
esac
41+
2242
LOG="experiments/logs/faster_rcnn_alt_opt_${NET}_${EXTRA_ARGS_SLUG}.txt.`date +'%Y-%m-%d_%H-%M-%S'`"
2343
exec&>>(tee -a"$LOG")
2444
echo Logging output to"$LOG"
2545

2646
time ./tools/train_faster_rcnn_alt_opt.py --gpu${GPU_ID} \
2747
--net_name${NET} \
2848
--weights data/imagenet_models/${NET}.v2.caffemodel \
29-
--imdbvoc_2007_trainval \
49+
--imdb${TRAIN_IMDB} \
3050
--cfg experiments/cfgs/faster_rcnn_alt_opt.yml \
3151
${EXTRA_ARGS}
3252

@@ -35,8 +55,8 @@ NET_FINAL=`grep "Final model:" ${LOG} | awk '{print $3}'`
3555
set -x
3656

3757
time ./tools/test_net.py --gpu${GPU_ID} \
38-
--def models/${NET}/faster_rcnn_alt_opt/faster_rcnn_test.pt \
58+
--def models/${PT_DIR}/${NET}/faster_rcnn_alt_opt/faster_rcnn_test.pt \
3959
--net${NET_FINAL} \
40-
--imdbvoc_2007_test \
60+
--imdb${TEST_IMDB} \
4161
--cfg experiments/cfgs/faster_rcnn_alt_opt.yml \
4262
${EXTRA_ARGS}
Lines changed: 35 additions & 15 deletions
Original file line numberDiff line numberDiff line change
@@ -1,9 +1,11 @@
11
#!/bin/bash
22
# Usage:
3-
# ./experiments/scripts/default_faster_rcnn.sh GPU NET [--set ...]
3+
# ./experiments/scripts/faster_rcnn_end2end.sh GPU NET DATASET [options args to {train,test}_net.py]
4+
# DATASET is either pascal_voc or coco.
5+
#
46
# Example:
5-
# ./experiments/scripts/default_faster_rcnn.sh 0ZF \
6-
# --set EXP_DIR foobar RNG_SEED 42 TRAIN.SCALES "[400,500,600,700]"
7+
# ./experiments/scripts/faster_rcnn_end2end.sh 0VGG_CNN_M_1024 pascal_voc \
8+
# --set EXP_DIR foobar RNG_SEED 42 TRAIN.SCALES "[400,500,600,700]"
79

810
set -x
911
set -e
@@ -13,25 +15,43 @@ export PYTHONUNBUFFERED="True"
1315
GPU_ID=$1
1416
NET=$2
1517
NET_lc=${NET,,}
16-
ITERS=70000
17-
DATASET_TRAIN=voc_2007_trainval
18-
DATASET_TEST=voc_2007_test
18+
DATASET=$3
1919

2020
array=($@ )
2121
len=${#array[@]}
22-
EXTRA_ARGS=${array[@]:2:$len}
22+
EXTRA_ARGS=${array[@]:3:$len}
2323
EXTRA_ARGS_SLUG=${EXTRA_ARGS///_}
2424

25-
LOG="experiments/logs/faster_rcnn_${NET}_${EXTRA_ARGS_SLUG}.txt.`date +'%Y-%m-%d_%H-%M-%S'`"
25+
case$DATASETin
26+
pascal_voc)
27+
TRAIN_IMDB="voc_2007_trainval"
28+
TEST_IMDB="voc_2007_test"
29+
PT_DIR="pascal_voc"
30+
ITERS=70000
31+
;;
32+
coco)
33+
# This is a very long and slow training schedule
34+
# You can probably use fewer iterations and reduce the
35+
# time to the LR drop (set in the solver to 350,000 iterations).
36+
TRAIN_IMDB="coco_2014_train"
37+
TEST_IMDB="coco_2014_minival"
38+
PT_DIR="coco"
39+
ITERS=490000
40+
;;
41+
*)
42+
echo"No dataset given"
43+
exit
44+
;;
45+
esac
46+
47+
LOG="experiments/logs/faster_rcnn_end2end_${NET}_${EXTRA_ARGS_SLUG}.txt.`date +'%Y-%m-%d_%H-%M-%S'`"
2648
exec&>>(tee -a"$LOG")
2749
echo Logging output to"$LOG"
2850

29-
NET_INIT=data/imagenet_models/${NET}.v2.caffemodel
30-
3151
time ./tools/train_net.py --gpu${GPU_ID} \
32-
--solver models/${NET}/faster_rcnn_end2end/solver.prototxt \
33-
--weights${NET_INIT} \
34-
--imdb${DATASET_TRAIN} \
52+
--solver models/${PT_DIR}/${NET}/faster_rcnn_end2end/solver.prototxt \
53+
--weightsdata/imagenet_models/${NET}.v2.caffemodel \
54+
--imdb${TRAIN_IMDB} \
3555
--iters${ITERS} \
3656
--cfg experiments/cfgs/faster_rcnn_end2end.yml \
3757
${EXTRA_ARGS}
@@ -41,8 +61,8 @@ NET_FINAL=`grep -B 1 "done solving" ${LOG} | grep "Wrote snapshot" | awk '{print
4161
set -x
4262

4363
time ./tools/test_net.py --gpu${GPU_ID} \
44-
--def models/${NET}/faster_rcnn_end2end/test.prototxt \
64+
--def models/${PT_DIR}/${NET}/faster_rcnn_end2end/test.prototxt \
4565
--net${NET_FINAL} \
46-
--imdb${DATASET_TEST} \
66+
--imdb${TEST_IMDB} \
4767
--cfg experiments/cfgs/faster_rcnn_end2end.yml \
4868
${EXTRA_ARGS}

0 commit comments

Comments
 (0)
[8]ページ先頭
©2009-2025 Movatter.jp