Disk Feature Extractor #49
Closed
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
sarlinpe
requested changes
Feb 7, 2021
hloc/extractors/disk.py
Outdated
| raise KeyError('Incompatible weight file!') | ||
| self.model = _DISK(window=8, desc_dim=conf['desc-dim']) | ||
| self.model.load_state_dict(weights) | ||
| self.model.to(torch.device('cuda' if torch.cuda.is_available() else 'cpu')) |
There was a problem hiding this comment.
The device selection is already handled by extract_features
hloc/extractors/disk.py
Outdated
Comment on lines
16
to
23
| # we rescale image only in the preprocessing step. | ||
| # we only pad the image size to be the multiple of 16 in this model | ||
| # 'height': None, # rescaled height (px). If unspecified, image is not resized in height dimension | ||
| # 'width': None, # rescaled width (px). If unspecified, image is not resized in width dimension | ||
| 'window': 5, # NMS window size | ||
| 'n': None, # Maximum number of features to extract. If unspecified, the number is not limited | ||
| 'desc-dim': 128, # descriptor dimension. Needs to match the checkpoint value | ||
| 'mode': 'nms', # Whether to extract features using the non-maxima suppresion mode or through training-time grid sampling technique |
There was a problem hiding this comment.
It would be great to clean up the comments and make the key names consistent with the other extractors (see my comments on the R2D2 PR).
hloc/extractors/disk.py
Outdated
| img = F.pad(img, (0, x_pad, 0, y_pad)) | ||
| assert img.shape[1] == new_h and img.shape[2] == new_w, "Wrong Padding!" | ||
|
|
||
| with torch.no_grad(): |
There was a problem hiding this comment.
The parent function should already set no_grad
hloc/extractors/disk.py
Outdated
Comment on lines
70
to
74
| msg = ('Please use input size which is multiple of 16 (or ' | ||
| 'adjust the --height and --width flags to let this ' | ||
| 'script rescale it automatically). This is because ' | ||
| 'we internally use a U-Net with 4 downsampling ' | ||
| 'steps, each by a factor of 2, therefore 2^4=16.') |
There was a problem hiding this comment.
Maybe the error message should be changed now
There was a problem hiding this comment.
I would like to remove this try-except block as we explicitly pad the image to be the multiple of 16.
hloc/extractors/disk.py
Outdated
| kps_img_space = kps_crop_space # (2, N) | ||
| x = kps_crop_space[0, :] | ||
| y = kps_crop_space[1, :] | ||
| mask = (0 <= x) & (x < orig_w) & (0 <= y) & (y < orig_h) |
There was a problem hiding this comment.
Shouldn't it be x <= (orig_w - 1) and similarly for y?
There was a problem hiding this comment.
Yeah I think this one is more accurate. Thanks!
hloc/extractors/disk.py
Outdated
Comment on lines
86
to
88
| # we didn't rescale so no need for this | ||
| # kps_img_space, mask = image.to_image_coord(kps_crop_space) | ||
| kps_img_space = kps_crop_space # (2, N) |
Merged
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Based on code in the original disk repo