zlccccc / 3DVL_Codebase

[CVPR2022 Oral] 3DJCG: A Unified Framework for Joint Dense Captioning and Visual Grounding on 3D Point Clouds
Other
51 stars 5 forks source link

Other bugs #5

Closed Zhang-Jing-Xuan closed 1 year ago

Zhang-Jing-Xuan commented 2 years ago

I try to visualize heatmap and run command:

python scripts/joint_scripts/heatmap_visualize.py --folder 2022-07-09_10-30-31_3DJCG --scene_id scene0011_00 --use_multiview --use_normal

Then, in the terminal, I meet an error: relation In addition, I try to visualize caption, but I haven't got val.hdf5 file which is necessary to visualize caption (line 42 in scripts/joint_scripts/caption_visualize.py). Can you provide the link to this file?

Zhang-Jing-Xuan commented 2 years ago

I try to train the 3DJCG mode without multiview features:

python scripts/joint_scripts/train_3djcg.py --use_color --use_normal --use_topdown --num_graph_steps 0 --num_locals 20 --batch_size 10 --epoch 37 --gpu 1 --verbose 50 --val_step 1000 --lang_num_max 8 --coslr --lr 0.002 --num_ground_epoch 0 --tag 3djcg
python scripts/joint_scripts/train_3djcg.py  --use_normal --use_topdown --num_graph_steps 0 --num_locals 20 --batch_size 10 --epoch 37 --gpu 1 --verbose 50 --val_step 1000 --lang_num_max 8 --coslr --lr 0.002 --num_ground_epoch 0 --tag 3djcg

but there is an error in the terminal: error

zxc351200 commented 2 years ago

For data_dict['relation'] in heatmap_visualize.py,you can write these code in the relation module fcaa7484bac3bfcf0d405fab550bbd4

For val.hdf5, please refer to Scan2Cap README.md

For train the 3DJCG mode without multiview features, you can change the code in the relation module as you need or you can change self.use_obj_embedding = True to self.use_obj_embedding = False int the relation module. 850d7fd5b1de227c78e460443e12d38

Zhang-Jing-Xuan commented 1 year ago

ok. Thank you for your reply.