MS COCO 2014 minival
Model | mAP@[0.5:0.95] (Original) | mAP@[0.5:0.95] (ChainerCV) |
---|---|---|
Light-Head R-CNN ResNet101 | 39.6 % [1] / 40.0 % [2] | 39.3 % |
Detect objects in an given image. This demo downloads MS COCO pretrained model automatically if a pretrained model path is not given.
$ python demo.py [--gpu <gpu>] [--pretrained-model <model_path>] <image>.jpg
The evaluation can be conducted using chainercv/examples/detection/eval_coco.py
.
You can train the model with the following code.
Note that this code requires chainermn
module.
$ mpiexec -n <#gpu> python train_multi.py [--batch-size <batch_size>]
- Zeming Li et al. "Light-Head R-CNN: In Defense of Two-Stage Object Detector" ArXiv 2017
- Light-Head R-CNN