Awesome

Cascade-RCNN_Tensorflow

Abstract

This is a tensorflow re-implementation of Cascade R-CNN Delving into High Quality Object Detection .

This project is completed by YangXue and WangYashan.

Train on VOC 2007 trainval and test on VOC 2007 test (PS. This project also support coco training.)

Comparison

use_voc2012_metric

Stage	AP50	AP60	AP70	AP75	AP80	AP85	AP90	AP95
baseline	75.80	67.25	52.15	41.41	27.98	12.63	2.73	0.11
1+2+3	75.80	68.74	57.09	48.68	37.70	22.52	7.51	0.54
1+2	75.98	68.40	56.01	46.89	35.67	20.42	6.44	0.39
1	74.89	65.98	52.45	40.63	27.79	13.22	2.94	0.11
2	75.67	68.69	56.73	47.82	35.5	20.29	6.46	0.38
3	74.35	67.62	56.64	48.65	38.02	23.19	8.05	0.54

use_voc2007_metric

Stage	AP50	AP60	AP70	AP75	AP80	AP85	AP90	AP95
baseline	73.62	65.28	51.93	42.52	29.48	16.2	5.84	1.32
1+2+3	73.69	66.59	56.19	48.82	39.47	25.57	12.09	2.5
1+2	74.01	66.5	55.53	46.53	36.96	23.6	11.33	2.15
1	72.92	64.29	52.41	48.8	30.36	16	5.64	2.15
2	73.55	66.75	55.78	48.35	37.39	23.61	10.66	2.69
3	71.58	65.73	56.64	49.08	39.68	26.25	12.28	2.32

Requirements

1、tensorflow >= 1.2
2、cuda8.0
3、python2.7 (anaconda2 recommend)
4、opencv(cv2)

Download Model

1、please download resnet50_v1、resnet101_v1 pre-trained models on Imagenet, put it to $PATH_ROOT/data/pretrained_weights.
2、please download mobilenet_v2 pre-trained model on Imagenet, put it to $PATH_ROOT/data/pretrained_weights/mobilenet.
3、please download trained model by this project, put it to $PATH_ROOT/output/trained_weights.

Data Format

├── VOCdevkit
│   ├── VOCdevkit_train
│       ├── Annotation
│       ├── JPEGImages
│   ├── VOCdevkit_test
│       ├── Annotation
│       ├── JPEGImages

Compile

cd $PATH_ROOT/libs/box_utils/cython_utils
python setup.py build_ext --inplace

Demo

Select a configuration file in the folder ($PATH_ROOT/libs/configs/) and copy its contents into cfgs.py, then download the corresponding weights.

cd $PATH_ROOT/tools
python inference.py --data_dir='/PATH/TO/IMAGES/' 
                    --save_dir='/PATH/TO/SAVE/RESULTS/' 
                    --GPU='0'

Eval

cd $PATH_ROOT/tools
python eval.py --eval_imgs='/PATH/TO/IMAGES/'  
               --annotation_dir='/PATH/TO/TEST/ANNOTATION/'
               --GPU='0'

Train

1、If you want to train your own data, please note:

(1) Modify parameters (such as CLASS_NUM, DATASET_NAME, VERSION, etc.) in $PATH_ROOT/libs/configs/cfgs.py
(2) Add category information in $PATH_ROOT/libs/label_name_dict/lable_dict.py     
(3) Add data_name to line 76 of $PATH_ROOT/data/io/read_tfrecord.py

2、make tfrecord

cd $PATH_ROOT/data/io/  
python convert_data_to_tfrecord.py --VOC_dir='/PATH/TO/VOCdevkit/VOCdevkit_train/' 
                                   --xml_dir='Annotation'
                                   --image_dir='JPEGImages'
                                   --save_name='train' 
                                   --img_format='.jpg' 
                                   --dataset='pascal'

3、train

cd $PATH_ROOT/tools
python train.py

Tensorboard

cd $PATH_ROOT/output/summary
tensorboard --logdir=.

Reference

1、https://github.com/endernewton/tf-faster-rcnn
2、https://github.com/zengarden/light_head_rcnn
3、https://github.com/tensorflow/models/tree/master/research/object_detection