You can not select more than 25 topics
Topics must start with a chinese character,a letter or number, can include dashes ('-') and can be up to 35 characters long.
lincolnz9511
4ab309bfd2
|
2 years ago | |
|---|---|---|
| .. | ||
| configs | 2 years ago | |
| README.md | 2 years ago | |
| convert-pretrain-to-detectron2.py | 2 years ago | |
| train_net.py | 2 years ago | |
README.md
MoCo: Transferring to Detection
The train_net.py script reproduces the object detection experiments on Pascal VOC and COCO.
Instruction
-
Install detectron2.
-
Convert a pre-trained MoCo model to detectron2's format:
python3 convert-pretrain-to-detectron2.py input.pth.tar output.pkl -
Put dataset under "./datasets" directory,
following the directory structure
requried by detectron2. -
Run training:
python train_net.py --config-file configs/pascal_voc_R_50_C4_24k_moco.yaml \ --num-gpus 8 MODEL.WEIGHTS ./output.pkl
Results
Below are the results on Pascal VOC 2007 test, fine-tuned on 2007+2012 trainval for 24k iterations using Faster R-CNN with a R50-C4 backbone:
| pretrain | AP50 | AP | AP75 |
|---|---|---|---|
| ImageNet-1M, supervised | 81.3 | 53.5 | 58.8 |
| ImageNet-1M, MoCo v1, 200ep | 81.5 | 55.9 | 62.6 |
| ImageNet-1M, MoCo v2, 200ep | 82.4 | 57.0 | 63.6 |
| ImageNet-1M, MoCo v2, 800ep | 82.5 | 57.4 | 64.0 |
Note: These results are means of 5 trials. Variation on Pascal VOC is large: the std of AP50, AP, AP75 is expected to be 0.2, 0.2, 0.4 in most cases. We recommend to run 5 trials and compute means.
