File size: 4,690 Bytes
5c8ef86 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 |
## Prepare datasets
In our paper, we conduct experiments on three common-used datasets, including Ref-COCO, Ref-COCO+ and G-Ref.
### 1. COCO 2014
The data could be found at [here](https://cocodataset.org/#download). Please run the following commands to download.
```shell
# download
mkdir datasets && cd datasets
wget http://images.cocodataset.org/zips/train2014.zip
# unzip
unzip train2014.zip -d images/ && rm train2014.zip
```
### 2. Ref-COCO
The data could be found at [here](https://github.com/lichengunc/refer). Please run the following commands to download and convert.
```shell
# download
wget https://bvisionweb1.cs.unc.edu/licheng/referit/data/refcoco.zip
# unzip
unzip refcoco.zip && rm refcoco.zip
# convert
python ../tools/data_process.py --data_root . --output_dir . --dataset refcoco --split unc --generate_mask
# lmdb
python ../tools/folder2lmdb.py -j anns/refcoco/train.json -i images/train2014/ -m masks/refcoco -o lmdb/refcoco
python ../tools/folder2lmdb.py -j anns/refcoco/val.json -i images/train2014/ -m masks/refcoco -o lmdb/refcoco
python ../tools/folder2lmdb.py -j anns/refcoco/testA.json -i images/train2014/ -m masks/refcoco -o lmdb/refcoco
python ../tools/folder2lmdb.py -j anns/refcoco/testB.json -i images/train2014/ -m masks/refcoco -o lmdb/refcoco
# clean
rm -r refcoco
```
### 3. Ref-COCO+
The data could be found at [here](https://github.com/lichengunc/refer). Please run the following commands to download and convert.
```shell
# download
wget https://bvisionweb1.cs.unc.edu/licheng/referit/data/refcoco+.zip
# unzip
unzip refcoco+.zip && rm refcoco+.zip
# convert
python ../tools/data_process.py --data_root . --output_dir . --dataset refcoco+ --split unc --generate_mask
# lmdb
python ../tools/folder2lmdb.py -j anns/refcoco+/train.json -i images/train2014/ -m masks/refcoco+ -o lmdb/refcoco+
python ../tools/folder2lmdb.py -j anns/refcoco+/val.json -i images/train2014/ -m masks/refcoco+ -o lmdb/refcoco+
python ../tools/folder2lmdb.py -j anns/refcoco+/testA.json -i images/train2014/ -m masks/refcoco+ -o lmdb/refcoco+
python ../tools/folder2lmdb.py -j anns/refcoco+/testB.json -i images/train2014/ -m masks/refcoco+ -o lmdb/refcoco+
# clean
rm -r refcoco+
```
### 4. Ref-COCOg
The data could be found at [here](https://github.com/lichengunc/refer). Please run the following commands to download and convert.
(Note that we adopt two different splits of this dataset, 'umd' and 'google'.)
```shell
# download
wget https://bvisionweb1.cs.unc.edu/licheng/referit/data/refcocog.zip
# unzip
unzip refcocog.zip && rm refcocog.zip
# convert
python ../tools/data_process.py --data_root . --output_dir . --dataset refcocog --split umd --generate_mask # umd split
mv anns/refcocog anns/refcocog_u
mv masks/refcocog masks/refcocog_u
python ../tools/data_process.py --data_root . --output_dir . --dataset refcocog --split google --generate_mask # google split
mv anns/refcocog anns/refcocog_g
mv masks/refcocog masks/refcocog_g
# lmdb
python ../tools/folder2lmdb.py -j anns/refcocog_u/train.json -i images/train2014/ -m masks/refcocog_u -o lmdb/refcocog_u
python ../tools/folder2lmdb.py -j anns/refcocog_u/val.json -i images/train2014/ -m masks/refcocog_u -o lmdb/refcocog_u
python ../tools/folder2lmdb.py -j anns/refcocog_u/test.json -i images/train2014/ -m masks/refcocog_u -o lmdb/refcocog_u
python ../tools/folder2lmdb.py -j anns/refcocog_g/train.json -i images/train2014/ -m masks/refcocog_g -o lmdb/refcocog_g
python ../tools/folder2lmdb.py -j anns/refcocog_g/val.json -i images/train2014/ -m masks/refcocog_g -o lmdb/refcocog_g
rm -r refcocog
```
### 5. Datasets struture
After the above-mentioned commands, the strutre of the dataset folder should be like:
```none
datasets
├── anns
│ ├── refcoco
│ │ ├── xxx.json
│ ├── refcoco+
│ │ ├── xxx.json
│ ├── refcocog_g
│ │ ├── xxx.json
│ ├── refcocog_u
│ │ ├── xxx.json
├── images
│ ├── train2014
│ │ ├── xxx.jpg
├── lmdb
│ ├── refcoco
│ │ ├── xxx.lmdb
│ │ ├── xxx.lmdb-lock
│ ├── refcoco+
│ │ ├── xxx.lmdb
│ │ ├── xxx.lmdb-lock
│ ├── refcocog_g
│ │ ├── xxx.lmdb
│ │ ├── xxx.lmdb-lock
│ ├── refcocog_u
│ │ ├── xxx.lmdb
│ │ ├── xxx.lmdb-lock
├── masks
│ ├── refcoco
│ │ ├── xxx.png
│ ├── refcoco+
│ │ ├── xxx.png
│ ├── refcocog_g
│ │ ├── xxx.png
│ ├── refcocog_u
│ │ ├── xxx.png
``` |