Skip to content

Latest commit

 

History

History
88 lines (74 loc) · 3.46 KB

HAWPv3.md

File metadata and controls

88 lines (74 loc) · 3.46 KB
Raw

HAWPv3: Learning Wireframes via Self-Supervised Learning

The codes of HAWPv3 are placed in the directory of hawp/ssl.

Model Name Comments MD5
hawpv3-fdc5487a.pth Trained on the images of Wireframe dataset fdc5487a43e3d42f6b2addf79d8b930d
hawpv3-imagenet-03a84.pth Trained on 100k images of ImageNet dataset 03a8400e9474320f2b42973d1ba19487

Inference on your own images

  • Run the following command line to obtain wireframes from HAWPv3 model

    hawpv3-fdc5487a.pth 
      python -m hawp.ssl.predict --ckpt checkpoints/hawpv3-fdc5487a.pth \
      --threshold 0.05 \
      --img {filename.png}
    hawpv3-imagenet-03a84.pth 
      python -m hawp.ssl.predict --ckpt checkpoints/hawpv3-imagenet-03a84.pth \
      --threshold 0.05 \
      --img {filename.png}

  • A running example on the DTU-24 images

    python -m hawp.ssl.predict --ckpt checkpoints/hawpv3-imagenet-03a84.pth  \
  --threshold 0.05  \
  --img ~/datasets/DTU/scan24/image/*.png \ 
  --saveto docs/figures/dtu-24 --ext png \

Training

python -m hawp.ssl.train --help
usage: train.py [-h] --datacfg DATACFG --modelcfg MODELCFG --name NAME
                [--pretrained PRETRAINED] [--overwrite] [--tf32]
                [--dtm {True,False}] [--batch-size BATCH_SIZE]
                [--num-workers NUM_WORKERS] [--base-lr BASE_LR]
                [--steps STEPS [STEPS ...]] [--gamma GAMMA] [--epochs EPOCHS]
                [--seed SEED] [--iterations ITERATIONS]

optional arguments:
  -h, --help            show this help message and exit
  --datacfg DATACFG     filepath of the data config
  --modelcfg MODELCFG   filepath of the model config
  --name NAME           the name of experiment
  --pretrained PRETRAINED
                        the pretrained model
  --overwrite           [Caution!] the option to overwrite an existed
                        experiment
  --tf32                toggle on the TF32 of pytorch
  --dtm {True,False}    toggle the deterministic option of CUDNN. This option
                        will affect the replication of experiments

training recipe:
  --batch-size BATCH_SIZE
                        the batch size of training
  --num-workers NUM_WORKERS
                        the number of workers for training
  --base-lr BASE_LR     the initial learning rate
  --steps STEPS [STEPS ...]
                        the steps of the scheduler
  --gamma GAMMA         the lr decay factor
  --epochs EPOCHS       the number of epochs for training
  --seed SEED           the random seed for training
  --iterations ITERATIONS
                        the number of training iterations