To do so, you just need to increase the upper-limit on the scale range by replacing -min-size 256 -max-size 1024 with -min-size 0 -max-size 9999 -min-scale 0.3 -max-scale 1.0. Note2: You can significantly improve the score (by ~4 pts) if you can afford more computations. Run python extract.py -help for more information.īy default, they corespond to what is used in the paper, i.e., a scale factor equal to 2^0.25 ( -scale-f 1.189207) and image size in the range ( -min-size 256 -max-size 1024). Note: You can modify the extraction parameters (scale factor, scale range.). scores ( N): keypoint scores (the higher the better).descriptors ( N x 128): l2-normalized descriptors.Scale denotes here the patch diameters in pixels. keypoints ( N x 3): keypoint position (x, y and scale).The keypoint file is in the npz numpy format and contains 3 fields: r2d2 extension.įor example, they will be saved in imgs/2d2 for the sample command above. txt image list.įor each image, this will save the top-k keypoints in a file with the same path as the image and a. This also works for multiple images (separated by spaces) or a. Python extract.py -model models/r2d2_WASF_N16.pt -images imgs/brooklyn.png -top-k 5000 To extract keypoints for a given image, simply execute: Here is a table that summarizes the performance of each model: model name faster2d2_WASF_N8_big.pt: The Fast-R2D2 equivalent of r2d2_WASF_N8.ptįor more details about the training data, see the dedicated section below.faster2d2_WASF_N16.pt: The Fast-R2D2 equivalent of r2d2_WASF_N16.pt.This can be interesting for certain applications like visual localization, but it implies a drop in MMA since keypoints gets slighlty less reliable. In other words, it outputs a higher density of keypoints. r2d2_WASF_N8_big.pt: Same than previous model, but trained with N=8 instead of N=16 in the repeatability loss.r2d2_WASF_N16.pt: this is the model used in the visual localization experiments (on HPatches It was trained with Web images ( W), Aachen day-time images ( A), Aachen day-night synthetic pairs ( S), and Aachen optical flow pairs ( F).r2d2_WAF_N16.pt: this is the model used in most experiments of the paper (on HPatches It was trained with Web images ( W), Aachen day-time images ( A) and Aachen optical flow pairs ( F).Conda install python tqdm pillow numpy matplotlib scipyĬonda install pytorch torchvision cudatoolkit=10.1 -c pytorch Pretrained modelsįor your convenience, we provide five pre-trained models in the models/ folder:
0 Comments
Leave a Reply. |