File size: 2,809 Bytes
36c95ba |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 |
Homography Regression by Gradient Descent
=========================================
This examples show how to use the `HomographyWarper` in order to do a regression where the parameter to optimize in this case is the homography driven by the gradient from a photometric loss.
Quick overview
--------------
.. code:: python
# load the data
img_src = load_image(os.path.join(args.input_dir, 'img1.ppm'))
img_dst = load_image(os.path.join(args.input_dir, 'img2.ppm'))
# instantiate the homography warper from `kornia`
height, width = img_src.shape[-2:]
warper = dgm.HomographyWarper(height, width)
# create the homography as the parameter to be optimized
# NOTE: MyHomography is an inherited nn.Module class
dst_homo_src = MyHomography().to(device)
# create optimizer
optimizer = optim.Adam(dst_homo_src.parameters(), lr=args.lr)
# main training loop
for iter_idx in range(args.num_iterations):
# send data to device
img_src, img_dst = img_src.to(device), img_dst.to(device)
# warp the reference image to the destiny with current homography
img_src_to_dst = warper(img_src, dst_homo_src())
# compute the photometric loss
loss = F.l1_loss(img_src_to_dst, img_dst)
# compute gradient and update optimizer parameters
optimizer.zero_grad()
loss.backward()
optimizer.step()
Downloading the data
====================
You can download the data by running: ``./download_data.sh``
Usage
=====
1. From root, run the docker development or build first if needed: ``//kornia/dev_en.sh``
2. Browse to ``cd /code/kornia/examples/homography_regression``
3. Install the dependencies by running: ``./install_dependencies.sh``
4. Now you can run the example followingthe instructions below:
.. code:: bash
python main.py --input-dir ./data --output-dir ./out --num-iterations 1000 --log-interval-vis 200 --cuda --lr 1e-3
.. code:: bash
main.py [-h] --input-dir INPUT_DIR --output-dir OUTPUT_DIR
[--num-iterations N] [--lr LR] [--momentum M] [--cuda]
[--seed S] [--log-interval N]
Homography Regression with perception loss.
optional arguments:
-h, --help show this help message and exit
--input-dir INPUT_DIR
the path to the directory with the input data.
--output-dir OUTPUT_DIR
the path to output the results.
--num-iterations N number of training iterations (default: 1000)
--lr LR learning rate (default: 1e-3)
--cuda enables CUDA training
--seed S random seed (default: 666)
--log-interval N how many batches to wait before logging training
status
--log-interval-vis N how many batches to wait before visual logging
training status
|