代码拉取完成,页面将自动刷新
Implementation of ADNet (https://sites.google.com/view/cvpr2017-adnet) in PyTorch 0.4.1.
References:
This implementation still cannot reproduce same performance with paper. Current performance distance (20px) precision with couple of test trials (each row is each trial):
SL | SL+RL | SL, nomd | SL+RL, nomd |
---|---|---|---|
75.3% | 73.3% | 54.9% | 56.5% |
70.7% | 69.2% | 59.4% | 47.1% |
69.0% | 71.0% | 55.3% | 49.3% |
68.7% | 72.9% | 54.5% | 57.6% |
75.5% | 68.9% | 54.5% | |
69.4% |
SL: Supervised Learning. RL: Reinforcement Learning, nomd: without multi-domain training
Inputs are welcome (especially RL part). Also currently cannot use multiple GPU (set CUDA_VISIBLE_DEVICES environment variable to select one GPU, if there are multiple GPUs in the system)
TODO:
Requirements:
├── datasets
(wrapper for datasets)
├── data
├── otb
├── Basketball
├── img
├── ....jpg
├── groundtruth_rect.txt
├── Biker
├── img
├── ....jpg
├── groundtruth_rect.txt
├── (continue till last class)
├── vot13
├── bicycle
├── ....jpg
├── camera_motion.label
├── groundtruth.txt
├── (and other files)
├── bolt
├── (similar structure with bicycle)
├── (continue till last class)
├── vot14
├── (similar structure with vot13)
├── vot15
├── (similar structure with vot13)
├── mains
(main.py for various datasets and tasks)
├── result_on_test_images
├── weights
├── (various main python files)
├── models
(network descriptions)
├── options
(template parser for command line inputs)
├── trainers
(define how model's forward / backward and logs)
├── weights
(the saved trained weights)
├── utils
(toolbox for drawing and scheduling)
├── videolist
├── vot13-otb.txt
├── vot14-otb.txt
├── vot15-otb.txt
├── (various python files)
├── scripts
(scripts for replicating experiement results)
├── work
(default folder to store logs/models)
├── vggm.pth
(the vggm weights for the base network)
python mains/ADNet.py --visualize True
ADNet_test
python ADNet_test.py --save_result_images results_on_test_images --display_images False
ADNet_ratiosamples_0.7
python ADNet_test.py --save_result_images results_on_test_images --display_images False --pos_samples_ratio 0.7
python create_plots.py --bboxes_folder results_on_test_images/ADNet_RL_-0.5 --show_plot False --save_plot_folder results_on_test_images/ADNet_RL_-0.5
此处可能存在不合适展示的内容,页面不予展示。您可通过相关编辑功能自查并修改。
如您确认内容无涉及 不当用语 / 纯广告导流 / 暴力 / 低俗色情 / 侵权 / 盗版 / 虚假 / 无价值内容或违法国家有关法律法规的内容,可点击提交进行申诉,我们将尽快为您处理。