wffancy / 3dssg Goto Github PK

Hi! Thanks for your great work. I'm a student in Beijing and I wanna reproduce your work. I saw the code link in 《3D scene graph prediction from point clouds》, but the running result is inconsistent with the article, what is the reason?

Hello,

thank you for the great work!

I ran the python scrpits/train.py code to train the model you provided, but I got an error that there is no dataset.

I downloaded the 3DSSG_subset folder from another issue, put it in the data folder ( /3dssg/data/3RScan/3DSSG_subset/ ), and set CONF.PATH.BASE in lib/config.py to my folder.

How can I download and use the data needed for train?

when I run on gpu， there is an error:

how can I fix this without changing the code

Hi.
Can you tell me the hyperparameter you use, I train the model but it not convergence.

Hello, thanks for your great work.

I setup this project following your guide and use the provided pointnet_cls_best_model.pth as pre-trained model to train 100 epochs. Through the provided visualization code, I found my model cannot predict the relationships. According to the visualization code, the predicted relationships that match GT should be shown as green, and those that do not should be black, as shown in the guide of this work.

But all of my visualized results shows the relationships as None:

And this is my logs (the last section), it seems the results are extremely poor:

------------------------iter: [100: 385200/385200]-------------------------
[loss] train_loss: 0.9935
[loss] train_obj_loss: 4.70025
[loss] train_pred_loss: 0.52347
[info] mean_fetch_time: 0.00075s       mean_forward_time: 0.10102s
[info] mean_backward_time: 0.18353s    mean_iter_time: 0.2853s
[info] ETA: 5h 58m 47s

saving last models...

training completed...

---------------------------------best-----------------------------
8----
[best] epoch: 1
[loss] loss: 1.01229
[loss] obj_loss: 4.88824
[loss] pred_loss: 0.52347
[sco.] Recall@5_object: 0.32947
[sco.] Recall@10_object: 0.46283
[sco.] Recall@3_predicate: 0.73218
[sco.] Recall@5_predicate: 0.73379
[sco.] Recall@50_relationship: 0.17119
[sco.] Recall@100_relationship: 0.18893

saving checkpoint...

saving last models...

Have I done anything wrong? How can I correctly predict the relationships? Looking forward to your reply. Thanks in advance.

Hi! Thanks for your great work. I'm a student in Beijing and I wanna reproduce your work. But I have a question that where can I get the files "relationships_train.json" and "relationships_validation.json"? Are they downloaded from somewhere or split by your own code? (I have downloaded the 3DSSG data-set, which only includes files named "affordances.txt" "attributes.txt" "classes.txt" "objects.json" "relationships.json" "relationships.txt" "wordnet_attributes.txt").
I would appreciate it if you could reply.

Hello, Thanks for your code first.

I hope to get scene graph file(objects.json, relationships.json) for new data. But I think I can get only visualization file with --vis and --use_pretrained. Do I need to correct code for JSON file?

I was able to get your program to work, I tried visualize(data_dict, model, obj_class_dict, pred_class_dict) using model_last.pth. obj_pred_cls is all set to 88 and There is nothing drawn in the prediction of the pdf file that is saved in the vis folder. I have set a breakpoint in vscode debug mode, but the program does not stop once in the following program. Am I forgetting some step to make it work correctly?

if i[j] >= 0.5:
pred_list.append(rel_pairs[index] + [j])
s, o = rel_pairs[index].
if s == o or j == 0:
continue
g1.edge(str(s.item()) + '', str(o.item()) + '', pred_dict[j])

Translated with www.DeepL.com/Translator (free version)

Hi, thanks for your work. I found, probably, a mistake on your codes.

The pooling section in gcn.py is not consistent with the Equation 4 in the 3DSSG paper. What you did is accumulating the own features based on the degree of that certain node. I demonstrated it on colab: https://colab.research.google.com/drive/1vBmdTxMfL3wKdm2mckjPZcZZVdfXTQjL?usp=sharing

And here is the official aggregation codes from Wald's cohort work (I don't know whether it is consistent with the paper either)
https://github.com/ShunChengWu/3DSSG/blob/69ad0818daaf7229fa193aeff4a608a34c189139/src/network_TripletGCN.py#LL70