-
Notifications
You must be signed in to change notification settings - Fork 13
Add some missing references #117
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
… zhou2023gloss, jiao2024visual
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thank you very much for your contribution, happy to see you join the effort :)
I left a few comments for minor changes.
For future PRs, I find it easier to review PRs that add a single paragraph / refer to a single topic (with a few citations). It makes it so I can review even if I have less time, and merge some changes while waiting for changes for others (just a preference).
src/references.bib
Outdated
@@ -1546,6 +1546,14 @@ @article{jiang2021sign | |||
year = {2021} | |||
} | |||
|
|||
@inproceedings{jiao2023cosign, | |||
title={CoSign: Exploring co-occurrence signals in skeleton-based continuous sign language recognition}, |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
need to add {}
to terms such as {CoSign}
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks for the suggestion, I have revised this reference item.
src/index.md
Outdated
@@ -538,10 +538,9 @@ Though some previous works have referred to this as "sign language translation," | |||
without handling the syntax and morphology of the signed language [@padden1988interaction] to create a spoken language output. | |||
Instead, SLR has often been used as an intermediate step during translation to produce glosses from signed language videos. | |||
|
|||
@jiang2021sign proposed a novel Skeleton Aware Multi-modal Framework with a Global Ensemble Model (GEM) for isolated SLR (SAM-SLR-v2) to learn and fuse multimodal feature representations. Specifically, they use a Sign Language Graph Convolution Network (SL-GCN) to model the embedded dynamics of skeleton keypoints and a Separable Spatial-Temporal Convolution Network (SSTCN) to exploit skeleton features. The proposed late-fusion GEM fuses the skeleton-based predictions with other RGB and depth-based modalities to provide global information and make an accurate SLR prediction. | |||
@jiang2021sign propose a novel Skeleton Aware Multi-modal Framework with a Global Ensemble Model (GEM) for isolated SLR (SAM-SLR-v2) to learn and fuse multimodal feature representations. Specifically, they use a Sign Language Graph Convolution Network (SL-GCN) to model the embedded dynamics of skeleton keypoints and a Separable Spatial-Temporal Convolution Network (SSTCN) to exploit skeleton features. The proposed late-fusion GEM fuses the skeleton-based predictions with other RGB and depth-based modalities to provide global information and make an accurate SLR prediction. @jiao2023cosign explore co-occurence signals in skeleton data to better exploit the knowledge of each signal for continuous SLR. Specifically, they use Group-specific GCN to abstract skeleton features from co-occurence signals (Body, Hand, Mouth and Hand) and introduce complementary regularization to ensure consistency between predictions based on two complementary subsets of signals. Additionally, they propose a two-stream framework to fuse static and dynamic information. The model demonstrates competitive performance cpmpared to video-to-gloss methods on the RWTH-PHOENIX-Weather-2014 [@koller2015ContinuousSLR], RWTH-PHOENIX-Weather-2014T [@cihan2018neural] and CSL-Daily [@dataset:Zhou2021_SignBackTranslation_CSLDaily] datasets. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
to minimize the diff, and for organization in more than one line, please add a new line before @jiao2023cosign (it will still show in one paragraph). I'd even propose to add a new line after every end of sentence, to make it easier to give comments
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
(but this paragraph looks good to me otherwise!)
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I have divided this paragraph into individual sentences to more clearly highlight the distinctions. Besides, in the previous version, I changed the tense of the previous sentence from past tense to present tense (@jiang2021sign proposed -> @jiang2021sign propose), and I recover the original version in the updated version.
Currently, it appears that the tenses in this project are not consistent and may require an overall review and correction.
src/index.md
Outdated
@@ -587,7 +586,11 @@ For this recognition, @cui2017recurrent constructs a three-step optimization mod | |||
First, they train a video-to-gloss end-to-end model, where they encode the video using a spatio-temporal CNN encoder | |||
and predict the gloss using a Connectionist Temporal Classification (CTC) [@graves2006connectionist]. | |||
Then, from the CTC alignment and category proposal, they encode each gloss-level segment independently, trained to predict the gloss category, | |||
and use this gloss video segments encoding to optimize the sequence learning model. | |||
and use this gloss video segments encoding to optimize the sequence learning model. @cheng2020fully propose a fully convolutional networks for continuous SLR, |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
(here as well, new line before sentence)
src/index.md
Outdated
@@ -587,7 +586,11 @@ For this recognition, @cui2017recurrent constructs a three-step optimization mod | |||
First, they train a video-to-gloss end-to-end model, where they encode the video using a spatio-temporal CNN encoder | |||
and predict the gloss using a Connectionist Temporal Classification (CTC) [@graves2006connectionist]. | |||
Then, from the CTC alignment and category proposal, they encode each gloss-level segment independently, trained to predict the gloss category, | |||
and use this gloss video segments encoding to optimize the sequence learning model. | |||
and use this gloss video segments encoding to optimize the sequence learning model. @cheng2020fully propose a fully convolutional networks for continuous SLR, |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
"a fully convolutional networks" should be "fully convolutional networks" or "a fully convolutional network"
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks for the suggestion, I have revised this sentence.
src/index.md
Outdated
@@ -587,7 +586,11 @@ For this recognition, @cui2017recurrent constructs a three-step optimization mod | |||
First, they train a video-to-gloss end-to-end model, where they encode the video using a spatio-temporal CNN encoder | |||
and predict the gloss using a Connectionist Temporal Classification (CTC) [@graves2006connectionist]. | |||
Then, from the CTC alignment and category proposal, they encode each gloss-level segment independently, trained to predict the gloss category, | |||
and use this gloss video segments encoding to optimize the sequence learning model. | |||
and use this gloss video segments encoding to optimize the sequence learning model. @cheng2020fully propose a fully convolutional networks for continuous SLR, | |||
moving away from LSTM-based methods to achieve end-to-end learning. They introduce a gloss feature enhancement (GFE) module to provide additional rectified supervision and |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
gloss feature enhancement should be capitalized (Gloss Feature Enhancement) because an acronym is introduced
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks for the suggestion, I have revised this sentence.
src/index.md
Outdated
and use this gloss video segments encoding to optimize the sequence learning model. @cheng2020fully propose a fully convolutional networks for continuous SLR, | ||
moving away from LSTM-based methods to achieve end-to-end learning. They introduce a gloss feature enhancement (GFE) module to provide additional rectified supervision and | ||
accelerate the training process. @min2021visual attribute the success of iterative training to its ability to reduce overfitting. They propose visual enhancement | ||
constraint (VEC) and visual alignment constraint (VAC) to strengthen the visual extractor and align long- and short-term predictions, enabling LSTM-based methods to be trained in an end-to-end manner. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
"visual enhancement constraint" should be capitalized, same for "visual alignment constraint"
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks for the suggestion, I have capitalized them.
src/index.md
Outdated
@@ -742,6 +745,10 @@ The model features shared representations for different modalities such as text | |||
on several tasks such as video-to-gloss, gloss-to-text, and video-to-text. | |||
The approach allows leveraging external data such as parallel data for spoken language machine translation. | |||
|
|||
@zhou2023gloss propose the GFSLT-VLP framework for gloss-free sign language translation, which improves SLT performance through visual-alignment pretraining. In the pretraining stage, they design a pretext task that aligns visual and textual |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
better imo from
@zhou2023gloss propose the GFSLT-VLP framework for gloss-free sign language translation, which improves SLT performance through visual-alignment pretraining.
to
@zhou2023gloss propose the Gloss-Free Sign Language Translation with Visual Alignment Pretraining (GFSLT-VLP) framework, to improve SLT performance.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks for the suggestion, I have revised this sentence.
src/index.md
Outdated
@@ -792,6 +798,10 @@ and showed similar performance, with the transformer underperforming on the vali | |||
They experimented with various normalization schemes, mainly subtracting the mean and dividing by the standard deviation of every individual keypoint | |||
either concerning the entire frame or the relevant "object" (Body, Face, and Hand). | |||
|
|||
@jiao2024visual propose a visual alignment pre-training framework for gloss-free sign language translation. Specifically, they adopt Cosign-1s [@jiao2023cosign] to obtain skeleton features from estimated pose sequences |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Cosign
or CoSign
?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
CoSign, thanks!
src/index.md
Outdated
@@ -792,6 +798,10 @@ and showed similar performance, with the transformer underperforming on the vali | |||
They experimented with various normalization schemes, mainly subtracting the mean and dividing by the standard deviation of every individual keypoint | |||
either concerning the entire frame or the relevant "object" (Body, Face, and Hand). | |||
|
|||
@jiao2024visual propose a visual alignment pre-training framework for gloss-free sign language translation. Specifically, they adopt Cosign-1s [@jiao2023cosign] to obtain skeleton features from estimated pose sequences | |||
and a pretrained text encoder to obtain corresponding textual features. During pretraining, these visual and textual features are aligned in a greedy manner. In the finetuning stage, they replace the shallow translation module | |||
used in pretraining with a pretrained translation module. This skeleton-based approach achieves state-of-the-art results on the RWTH-PHOENIX-Weather-2014T [@cihan2018neural], CSL-Daily [@dataset:Zhou2021_SignBackTranslation_CSLDaily], OpenASL [@shi-etal-2022-open], and How2Sign[@dataset:duarte2020how2sign] datasets without relying on gloss annotations. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
missing space after How2Sign
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The space has been added.
Thank you for your comprehensive feedback. I have revised relevant parts and restructured the sentences for clarity. Additionally, I’ve discovered that incorporating hyperlinks for each reference can greatly enhance the document’s usability. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Great job!
Added missing references: cheng2020fully, min2021visual, jiao2023cosign, zhou2023gloss, jiao2024visual.
Plan to include more references from VIPL-SLP/awesome-sign-language-processing in future updates.