Now it's training on a K80, iter 24k
- Current mAP: 0.751 (Expected 0.772)
- Current iteration number: 45k
- Train from pretrained vgg weight
- With data augmentation
- Training dataset: Pascal 07 + 12
- Val dataset: Pascal 07test
- Train log: res.json
- Update config.py, set paths
- Command line:
python train.py
- You can download pretrained vgg weight here.
- You can download annotation file in json format here
- You can download Pascal dataset here
- It's still dirty code, wait to refine
- https://github.com/fastai/fastai
- https://github.com/amdegroot/ssd.pytorch
- Liu, Wei, et al. "SSD: Single shot multibox detector." European conference on computer vision. Springer, Cham, 2016.
- Training dataset only
- Exponentially weighted average