Closed sunanhe closed 2 years ago
Our data preprocessing strategy is the same as Scan2Cap. In addition, our training process should be independent of Scan2Cad. We will update the code as soon as possible. You may refer to the official preprocessing codes in Scan2Cap. https://github.com/daveredrum/Scan2Cap
Thanks for your timely reply. I have a new problem.
Traceback (most recent call last):
File "scripts/joint_scripts/train_3djcg.py", line 503, in <module>
train(args)
File "scripts/joint_scripts/train_3djcg.py", line 433, in train
solver, num_params, root = get_solver(args, dataset, dataloader)
File "scripts/joint_scripts/train_3djcg.py", line 176, in get_solver
model = get_model(args, dataset["train"], device)
File "scripts/joint_scripts/train_3djcg.py", line 88, in get_model
dataset_config=DC
File "3DJCG/models/jointnet/jointnet.py", line 65, in __init__
self.caption = SceneCaptionModule(vocabulary, embeddings, emb_size, 128, caption_hidden_size, num_proposal)
File "3DJCG/models/capnet/caption_module.py", line 83, in __init__
raise NotImplementedError()
NotImplementedError
We use the model: _https://github.com/zlccccc/3DJCG/blob/main/models/capnet/caption_module.py#L86_ For the training scripts, please use: _python scripts/joint_scripts/train_3djcg.py --use_multiview --use_normal --use_topdown --num_graph_steps 0 --num_locals 20 --batch_size 10 --epoch 200 --gpu 2 --verbose 50 --val_step 1000 --lang_num_max 8 --coslr --lr 0.002 --num_groundepoch 150 --tag 3djcg We will update the code and README as soon as possible.
Hi, I'm really interested in this work. But where is ScanRefer_filtered_organized.json? I can't find it in ScanRefer dataset.
We are using the original ScanRefer dataset, the downloaded file name is ScanRefer_filtered_organized.json. Please follow the official link https://github.com/daveredrum/ScanRefer to fill out this form. Once your request is accepted, you will receive an email with the download link.
OK, I can run it now.
We did a double-check to make sure that the original ScanRefer & Scan2Cap datasets were used for both training and testing in the paper. It is possible that we changed the dataset name when we downloaded it earlier. Please change "ScanRefer_filtered_organized.json" to "ScanRefer_filtered.json" in the code.
ScanRefer_filtered_organized.json can be obtained by [Code].
Hi, I'm really interested in this work. But I cannot follow the Data preparation. It seems lack of the instruction for SCAN2CAD dataset.