Update citation of Seq2Seq-NMS

docs/reference/object-detection-2d-nms-seq2seq_nms.md

@@ -5,7 +5,7 @@ The *seq2seq-nms* module contains the *Seq2SeqNMSLearner* class, which inherits
 ### Class Seq2SeqNMSLearner
 Bases: `engine.learners.Learner`
 
-It can be used to perform single-class non-maximum suppression (NMS) on images (inference) as well as training new seq2seq-nms models. The implementation is based on [[1]](#seq2seq_nms-1). The method is set-up for performing NMS on the person-detection task, using the implemention of the [SSD](/docs/reference/object-detection-2d-ssd.md) detector. The Seq2Seq-NMS method can also be employed for performing single-class NMS, in any class other than human/pedestrian class. In that case the method needs to be trained from scratch. Finally, a pretrained-model can be employed for evaluation or inference on the same class that it was trained with, using RoIs from a different detector than the one used in the training. In that case, we advise to fine-tune the Seq2Seq-nms pretrained model using RoIs from the detector, deployed in the inference/evaluation of the method, in order to achieve the highest possible performance.
+It can be used to perform single-class non-maximum suppression (NMS) on images (inference) as well as training new seq2seq-nms models. The implementation is based on [1]. The method is set-up for performing NMS on the person-detection task, using the implemention of the [SSD](/docs/reference/object-detection-2d-ssd.md) detector. The Seq2Seq-NMS method can also be employed for performing single-class NMS, in any class other than human/pedestrian class. In that case the method needs to be trained from scratch. Finally, a pretrained-model can be employed for evaluation or inference on the same class that it was trained with, using RoIs from a different detector than the one used in the training. In that case, we advise to fine-tune the Seq2Seq-nms pretrained model using RoIs from the detector, deployed in the inference/evaluation of the method, in order to achieve the highest possible performance.
 
 The [Seq2SeqNMSLearner](/src/opendr/perception/object_detection_2d/nms/seq2seq_nms/seq2seq_nms_learner.py) class has the following
 public methods:
@@ -302,4 +302,4 @@ TABLE-1: Average Precision (AP) achieved by pretrained models on the person dete
 
 
 #### References
-<a name="seq2seq_nms-1" href="https://www.techrxiv.org/articles/preprint/Neural_Attention-driven_Non-Maximum_Suppression_for_Person_Detection/16940275">[1]</a> Neural Attention-driven Non-Maximum Suppression for Person Detection, [TechRxiv](https://www.techrxiv.org/articles/preprint/Neural_Attention-driven_Non-Maximum_Suppression_for_Person_Detection/16940275).
+[1] C. Symeonidis, I. Mademlis, I. Pitas and N. Nikolaidis, "Neural Attention-Driven Non-Maximum Suppression for Person Detection" in IEEE Transactions on Image Processing, vol. 32, pp. 2454-2467, 2023, doi: 10.1109/TIP.2023.3268561.

src/opendr/perception/object_detection_2d/nms/seq2seq_nms/README.md

@@ -1,7 +1,9 @@
 Seq2Seq-NMS
 ======
 
-This folder contains an implementation of Seq2Seq-NMS [[1]](#seq2seq_nms-1).
+This folder contains an implementation of the Seq2Seq-NMS algorithm, for neural Non-Maximum Suppression in visual person detection. If one uses any part of this implementation in his/her work, he/she is kindly asked to cite the following paper:
+
+- C. Symeonidis, I. Mademlis, I. Pitas and N. Nikolaidis, "Neural Attention-Driven Non-Maximum Suppression for Person Detection" in IEEE Transactions on Image Processing, vol. 32, pp. 2454-2467, 2023, doi: 10.1109/TIP.2023.3268561.
 
 TABLE-1: Average Precision (AP) achieved by pretrained models on the person detection task on the validation sets. The maximum number or RoIs, employed for the performance evaluation was set to 800.
 |  **Pretrained Model**  | **Dataset** | **Detector** | **Type of Appearance-based Features** | **Pre-processing IoU Threshold** | **AP@0.5 on validation set** | **AP@0.5 on test set** |
@@ -14,4 +16,3 @@ TABLE-1: Average Precision (AP) achieved by pretrained models on the person dete
 \* The minival set was used as validation set.<br>
 \*\* The minitest set was used as test set.
 
-<a name="seq2seq_nms-1" href="https://www.techrxiv.org/articles/preprint/Neural_Attention-driven_Non-Maximum_Suppression_for_Person_Detection/16940275">[1]</a> Neural Attention-driven Non-Maximum Suppression for Person Detection, [TechRxiv](https://www.techrxiv.org/articles/preprint/Neural_Attention-driven_Non-Maximum_Suppression_for_Person_Detection/16940275).