Skip to content

Commit

Permalink
update
Browse files Browse the repository at this point in the history
  • Loading branch information
ZhangGongjie committed Mar 15, 2022
1 parent a125d00 commit 0e00bda
Showing 1 changed file with 4 additions and 5 deletions.
9 changes: 4 additions & 5 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -13,7 +13,7 @@ attributed to the complication in matching object queries to encoded image featu
in DETR's decoder cross-attention modules.

<div align=center>
<img src='.assets/matching_complication.jpg' width="62.5%">
<img src='.assets/matching_complication.jpg' width="68%">
</div>

Motivated by this observation, in our paper, we propose SAM-DETR, a
Expand All @@ -35,15 +35,15 @@ reference box for each object query, whose center location is used to generate
corresponding position embeddings.

<div align=center>
<img src='.assets/decoder_layer.jpg' width="72.5%">
<img src='.assets/decoder_layer.jpg' width="86.5%">
</div>

The figure below illustrates the architecture of the appended "Semantics Aligner", which
aligns the semantics of "encoded image features" and "object queries" by resampling features
from multiple salient points as new object queries.

<div align=center>
<img src='.assets/semantics_aligner.jpg' width="65%">
<img src='.assets/semantics_aligner.jpg' width="75%">
</div>

Being like a plug-and-play, our approach can be
Expand Down Expand Up @@ -118,7 +118,7 @@ python test.py # unit test (should see all checking is True)
Please download [COCO 2017 dataset](https://cocodataset.org/) and organize them as following:

```
sam_detr_root/
code_root/
弩岸岸 data/
弩岸岸 coco/
念岸岸 train2017/
Expand All @@ -134,7 +134,6 @@ sam_detr_root/




## Usage

### Reproducing Paper Results
Expand Down

0 comments on commit 0e00bda

Please sign in to comment.