Comments (2)
Hi, thank you for the kind comment.
We have released all the best models for each set of local feature extractors tested and reported in our manuscript. For example, for SuperPoint we released weights for the two best OpenGlue models: one used SuperPoint with KITTI, and the other used weights trained on the COCO dataset.
For SIFT descriptors, on the other hand, you can find two models: rotation and with scale-rotation included during training. The combination of rotation and scale showed better results in our experiments, but the difference was not severe, so we decided to include both options.
from openglue.
I am getting very bad results with the coco version of openglue. I used this configuration :
name:
'SuperPointNetBn'
descriptor_dim:
&DIM 256
parameters:
max_keypoints:
2048
descriptor_dim:
*DIM
nms_kernel:
9
remove_borders_size:
4
keypoint_threshold:
0.005
weights:
'/home/ostap/projects/pytorch-superpoint/logs/superpoint_coco_heat2_0/checkpoints/superPointNet_170000_checkpoint.pth.tar`
For the kitti version, the results are better but very far from the matching produced by superGlue, I used this configuration :
data:
root_path:
'/datasets/extra_space2/ostap/MegaDepth'
train_list_path:
'/home/ostap/projects/superglue-lightning/assets/megadepth_train_2.0.txt'
val_list_path:
'/home/ostap/projects/superglue-lightning/assets/megadepth_valid_2.0.txt'
test_list_path:
'/home/ostap/projects/superglue-lightning/assets/megadepth_valid_2.0.txt'
batch_size_per_gpu:
2
dataloader_workers_per_gpu:
4
target_size:
[ 960, 720 ]
val_max_pairs_per_scene:
50
train_pairs_overlap:
[0.15, 0.7]
inference:
match_threshold:
0.2
superglue:
descriptor_dim:
&DESCRIPTOR_DIM 256
laf_to_sideinfo_method:
'none'
positional_encoding:
hidden_layers_sizes:
[32, 64, 128]
output_size:
*DESCRIPTOR_DIM
attention_gnn:
num_stages:
9
num_heads:
4
embed_dim:
*DESCRIPTOR_DIM
attention:
'softmax'
use_offset:
False
dustbin_score_init:
1.0
otp:
num_iters:
20
reg:
1.0
residual:
True
features:
name:
'SuperPointNetBn'
descriptor_dim:
&DIM 256
parameters:
max_keypoints:
1024
descriptor_dim:
*DIM
nms_kernel:
3
remove_borders_size:
4
keypoint_threshold:
0.005
weights:
'/home/hc/workspace/OpenGlue/experiment_1/superPointNet_50000_checkpoint.pth.tar'
Are these configurations the right ones ?
from openglue.
Related Issues (15)
- Training on custom dataset HOT 3
- The weight of the SuperPoint HOT 2
- How to generate the paris.txt HOT 2
- where is the .pth of the superglue,the originnal superglue .pth can't fit your code ,thank you
- Out of bounds error in run_nms
- another work of training your worn graph matcher HOT 1
- AUC is always 0 during trainning HOT 2
- The function draw_LAF_matches() not defined HOT 2
- /examples folder missing HOT 1
- Not working with SIFT instead of OPENCV_SIFT
- Typical Training Time HOT 4
- problem
- Info regarding number of keypoints for training
- Where is "Image-matching.ipynb"?
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from openglue.