tanghanlin / detectron-pytorch

Detectron-PYTORCH

Play with state-of-the-art pedestrian detectors.

Requirements

1. pytorch
2. tensorflow (only tensorboard is used)
3. tensorboardX

   pip install tensorboardX
   

4. visdom

   pip install visdom
   

Installation

1. Compiling libs

   cd ./libs
   make

2. citypersons and cityscape dataset: download coco and extract zips under ./data/citypersons/, so your dirs look like,

   ./data/
           data/citypersons/{annotations}
           data/citypersons/cityscape
               data/citypersons/cityscape/leftImg8bit/{train|val|test}
               data/citypersons/cityscape/gtFine/{train|val|test}

Create citypersons extended dataset

```bash
    python libs/datasets/citypersons2.py
```

Training and evaluation models

```bash
sh ./exper/citypersons/retinanet/train_test_resnet50_softmax.sh
```