GitHub - baituhuangyu/faster-rcnn-tf-demo: 快速搭建目标检测项目

Branches Tags

Name		Name	Last commit message	Last commit date
Latest commit History 4 Commits
anchor_generators		anchor_generators
box_coders		box_coders
builders		builders
core		core
data_decoders		data_decoders
dataset_tools		dataset_tools
deployment		deployment
inference		inference
legacy		legacy
matchers		matchers
meta_architectures		meta_architectures
metrics		metrics
models		models
nets		nets
predictors		predictors
protos		protos
sample		sample
utils		utils
.gitignore		.gitignore
README		README
eval_util.py		eval_util.py
export_inference_graph.py		export_inference_graph.py
exporter.py		exporter.py
inputs.py		inputs.py
model_hparams.py		model_hparams.py
model_lib.py		model_lib.py
model_main.py		model_main.py
train_main.sh		train_main.sh

Repository files navigation

# 修改

## 将TensorFlow 官方的代码修改，目录结构的修改，net从slim移植过来

# faster rcnn demo
- backbone pnasnet
  - todo
    - FPN类似的【多级特征提取】添加到rpn侧，
    - roipooling -> RoIAlign，
    - batch normal -> group normal，或者自适应
    - nms -> softer nms
    - 图像增强，data augmentation, 如mixup：BEYOND EMPIRICAL RISK MINIMIZATION


```
train_main.sh
sample/config/pipeline.config

dataset_tools/create_pascal_tf_record.py
dataset_tools/create_coco_dataset.sh
dataset_tools/create_coco_tf_record.py

data/*.json
data/*.pbtxt

```

# 需要添加backbone 时，在 `nets` 下添加
# 需要切换到ssd或者maskrcnn 时，在 `nets` 、 `meta_architectures`、 `models` 下添加