One class training for instance segmentation

[JackieXuu]

I just have one class, how to set config file? and we set unlabel points as -100 as your config, and 0 for semantic labels. Is it right?

The training is so weird for this. Can you help see it?

2023-03-15 11:20:55,744 - INFO - Training                                                                                                                                                                  
2023-03-15 11:21:09,408 - INFO - Epoch [1/128][10/45]  lr: 0.004, eta: 2:10:54, mem: 893, data_time: 0.00, iter_time: 0.90, semantic_loss: 0.0000, offset_loss: 1.0544, cls_loss: 0.2068, mask_loss: 0.0000
, iou_score_loss: 0.0000, num_pos: 0.0000, num_neg: 10.0000, loss: 1.2613                                                                                                                                  
2023-03-15 11:21:18,513 - INFO - Epoch [1/128][20/45]  lr: 0.004, eta: 1:48:53, mem: 901, data_time: 0.00, iter_time: 0.99, semantic_loss: 0.0000, offset_loss: 1.2640, cls_loss: 0.0839, mask_loss: 0.0000
, iou_score_loss: 0.0000, num_pos: 0.0000, num_neg: 8.0000, loss: 1.3479                                                                                                                                   
2023-03-15 11:21:27,877 - INFO - Epoch [1/128][30/45]  lr: 0.004, eta: 1:42:16, mem: 901, data_time: 0.00, iter_time: 0.85, semantic_loss: 0.0000, offset_loss: 1.4361, cls_loss: 0.0335, mask_loss: 0.0000
, iou_score_loss: 0.0000, num_pos: 0.0000, num_neg: 7.0000, loss: 1.4695                                                                                                                                   
2023-03-15 11:21:37,081 - INFO - Epoch [1/128][40/45]  lr: 0.004, eta: 1:38:30, mem: 907, data_time: 0.00, iter_time: 0.85, semantic_loss: 0.0000, offset_loss: 1.2537, cls_loss: 0.0160, mask_loss: 0.0000
, iou_score_loss: 0.0000, num_pos: 0.0000, num_neg: 14.0000, loss: 1.2698                                                                                                                                  
2023-03-15 11:21:42,295 - INFO - Validation                                                          
100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 45/45 [00:14<00:00,  3.17it/s]
2023-03-15 11:21:56,482 - INFO - Evaluate instance segmentation                                      
Traceback (most recent call last):                                                                                                                                                                         
  File "tools/train.py", line 206, in <module>                                                       
    main()                                                                                                                                                                                                 
  File "tools/train.py", line 201, in main                                                           
    validate(epoch, model, val_loader, cfg, logger, writer)                                                                                                                                                
  File "tools/train.py", line 121, in validate                                                                                                                                                             
    eval_res = scannet_eval.evaluate(all_pred_insts, all_gt_insts)                                                                                                                                         
  File "/home/a/SoftGroup/softgroup/evaluation/instance_eval.py", line 398, in evaluate                                                                                                            
    ap_scores, rc_scores = self.evaluate_matches(matches)                                                                                                                                                  
  File "/home/a/SoftGroup/softgroup/evaluation/instance_eval.py", line 161, in evaluate_matches                                                                                                    
    num_true_examples = y_true_sorted_cumsum[-1]                                                                                                                                                           
IndexError: index -1 is out of bounds for axis 0 with size 0    

[andou36]

@thangvubk @ZhengtianXu I have the same problem, how does softgroup work for single class instance segment, did you solve it?

[kaoutar-elmouh]

@thangvubk@ZhengtianXu J'ai le même problème, comment fonctionne le softgroup pour un segment d'instance à classe unique, l'avez-vous résolu ?

did you find th solution ?