Update code

rd20karim · Aug 21, 2024 · 2d402d6 · 2d402d6
1 parent d771bd9
commit 2d402d6
Show file tree

Hide file tree

Showing 3 changed files with 17 additions and 31 deletions.
diff --git a/README.md b/README.md
@@ -1,12 +1,13 @@
 ## Description
 
-Official implementation of our paper for interpretable motion to text generation:
+[BMVC 2024] Official implementation of our paper for interpretable motion to text generation:
 
 <div align="center">
 
-[<span style="font-size: 25px;">Guided Attention for Interpretable Motion Captioning</span>](https://hal.science/hal-04251363v1) 
+[<span style="font-size: 25px;">Guided Attention for Interpretable Motion Captioning (BMVC 2024)</span>](https://hal.science/hal-04251363v1) 
 
 [![arxiv](https://img.shields.io/badge/arXiv-Motion_to_Text-red?logo=arxiv)](https://arxiv.org/abs/2310.07324v1)
+[![BMVC](https://img.shields.io/badge/BMVC2024-gold)](https://bmvc2024.org/programme/accepted_papers/)
 [![License](https://img.shields.io/badge/License-MIT-green)]()
 
 </div>
@@ -20,13 +21,11 @@ Official implementation of our paper for interpretable motion to text generation
 If you find this code or paper useful in your work, please cite:
 
 ```
-@article{radouane2023guided,
+@INPROCEEDINGS{radouane2024guided,
       title={Guided Attention for Interpretable Motion Captioning}, 
       author={Karim Radouane and Andon Tchechmedjiev and Sylvie Ranwez and Julien Lagarde},
-      year={2023},
-      eprint={2310.07324},
-      archivePrefix={arXiv},
-      primaryClass={cs.CV}
+      booktitle = {Proceedings of the 35th British Machine Vision Conference},
+      year={2024}
 }
 ```
 
@@ -130,6 +129,7 @@ python nlg_evaluation.py
 Generate skeleton animation and attention maps (adaptive+spatio-temporal):
 
 ```
+set PYTHONPATH=Project_Absolute_Path
 python visualizations/visu_word_attentions.py --path PATH --n_map NUMBER_ATTENTION_MAP --n_gifs NUMBER_3D_ANIMATIONS --save_results DIRECTORY_SAVE_PLOTS
 ```
 

diff --git a/architectures/decode_beam_2.py b/architectures/decode_beam_2.py
@@ -31,8 +31,6 @@ def __init__(self, hiddenstate, previousNode, wordId, logProb, length,att_weight
         self.att_position = att_position
     def eval(self, alpha=1.0):
         reward = 0
-        # Add here a function for shaping a reward
-
         return self.logp / float(self.leng - 1 + 1e-6) + alpha * reward
 
 
@@ -50,18 +48,16 @@ def beam_decode(self,target_tensor, decoder_hiddens, encoder_outputs=None):
     decoded_batch = []
     dec_pred_output = []
 
-
     # decoding goes sentence by sentence
     target_tensor = target_tensor.permute(1,0)
     B = target_tensor.size(0) # batch_size
-    #TODO REMOVE DEBUG PARAM IN THE FINALE VESION
-    debug_param = B
+
     #  ------------ Spatial-Temp pose feature : R # shape  : (T,N,V,C)
     T, N, CV = self.x.size()
     R = encoder_outputs.reshape(T, N, 6 * self.hidden_size).to(self.device)
 
     # TODO CHANGE THIS CODE  TO DECODE PER BATCH
-    for idx in range(debug_param):
+    for idx in range(B):
         #x = xparts #shape[T,B,K,D] (k=6) (D=feature_dimension)
         x = self.x[:,idx,:].unsqueeze(1)
         enc_masks = torch.zeros((x.shape[0],1)) # (seq_len,batch_size)
@@ -100,7 +96,6 @@ def beam_decode(self,target_tensor, decoder_hiddens, encoder_outputs=None):
         qsize = 1
         # start beam search
         while True:
-        #for idx in range(trg_len-1):
             # give up when decoding takes too long
             if qsize > 100: break
             # fetch the best node
@@ -138,7 +133,7 @@ def beam_decode(self,target_tensor, decoder_hiddens, encoder_outputs=None):
 
             # ------ TEMPORAL ATTENTION shape : (T,N,V*C)--------------------------------------------------------
             ep_t = self.temporal_att(torch.tanh(
-                self.tempfeat_extract_hdec(bot_ht_1).expand(x.size()[:2] + (hidden_dim,)) + self.feat_extract_g(R)))  # +self.speed_temp(velocity)))
+                self.tempfeat_extract_hdec(bot_ht_1).expand(x.size()[:2] + (hidden_dim,)) + self.feat_extract_g(R)))
             ep_t = ep_t.masked_fill(enc_masks.unsqueeze(-1).to(self.device) == 0, float('-inf'))
             b_t = torch.softmax(ep_t, dim=0)
 
@@ -172,9 +167,6 @@ def beam_decode(self,target_tensor, decoder_hiddens, encoder_outputs=None):
             decoder_hidden = (bot_ht_1,bot_mt_1,top_ht_1,top_mt_1)
 
             """--------------------------------- END MODEL PREDICTION--------------------------------"""
-            #TODO ADAPT
-            # decoder_logits, decoder_hidden = self.dec(torch.tensor([[decoder_input]],device=self.device), decoder_hidden, encoder_output,enc_masks)
-
             decoder_output = torch.log_softmax(decoder_logits,axis=-1)
             # PUT HERE REAL BEAM SEARCH OF TOP
             log_prob, indexes = torch.topk(decoder_output.squeeze(0), beam_width)
@@ -213,10 +205,4 @@ def beam_decode(self,target_tensor, decoder_hiddens, encoder_outputs=None):
             utterances.append(utterance)
         decoded_batch.append(utterances)
 
-    return decoded_batch
-
-if __name__=="__main__":
-    hidden_size = 64
-    embedding_dim = 64
-    #decoder = seq2seq(642, hidden_size, embedding_dim, num_layers=1, device=device,
-    #                  bidirectional=False, attention="local", mask=True, beam_size=2).to(device)
+    return decoded_batch
diff --git a/src/evaluate_m2L.py b/src/evaluate_m2L.py
@@ -24,7 +24,7 @@ def load_model_config(args= None,device=None):
         path_txt = project_path+"\datasets\sentences_corrections.csv"
         path_motion = aug_path+"\kit_with_splits_2023.npz"
 
-        # -----------  H3D IMPORTS   ---------------------
+        # -----------  [H3D IMPORTS]   ---------------------
     elif args.dataset_name=="h3D":
         from architectures.LSTM_h3D import seq2seq
         from datasets.h3d_m2t_dataset_ import dataset_class
@@ -55,12 +55,12 @@ def evaluate(loaded_model,data_loader,mode,multiple_references=False,name_file=N
     loaded_model.eval()
     epoch_loss = 0
     output_per_batch,target_per_batch = [],[]
-    name_file = f"./Predictions/LSTM_{args.dataset_name}_{name_file}_{args.lambdas}"
-    BLEU_scores = []
-    with open(name_file + ".csv", mode="w") as _:pass
-    logging.info(f"Compute BLEU scores per batch and write predictions/refs to {name_file}")
     loaded_model.eval()
     if beam_size==1:
+        name_file = f"./Predictions/LSTM_{args.dataset_name}_{name_file}_{args.lambdas}"
+        BLEU_scores = []
+        with open(name_file + ".csv", mode="w") as _:pass
+        logging.info(f"Compute BLEU scores per batch and write predictions/refs to {name_file}")
         for i, batch in enumerate(data_loader):
             loss_b,bleu_score_4,pred,refs = run_batch(model=loaded_model,batch=batch,data_loader=data_loader,mode=mode,teacher_force_ratio=0,
                                                           device=device,multiple_references=multiple_references,attention_type=args.attention_type)
@@ -88,7 +88,7 @@ def evaluate(loaded_model,data_loader,mode,multiple_references=False,name_file=N
 
     else:
         logging.info("START BEAM SEARCHING")
-        file_save_beam = f"./Predictions/LSTM_{args.dataset_name}_preds_{args.lambdas}_beam_size_{beam_size}.csv"
+        file_save_beam = f"./Predictions/LSTM_{args.dataset_name}_{name_file}_{args.lambdas}_beam_size_{beam_size}.csv"
         with open(file_save_beam,'w'): pass #create the file
         beam_bleus = [[] for _ in range(beam_size)]
         for i, batch in enumerate(data_loader):