mkt1412 / graspgpt_public Goto Github PK
View Code? Open in Web Editor NEWcode implementation of GraspGPT and FoundationGrasp
code implementation of GraspGPT and FoundationGrasp
FileNotFoundError: [Errno 2] No such file or directory: 'gcngrasp/../data/taskgrasp/misc.pkl'
Hello! Thank you for your awesome work! I noticed that you train your network on a RTX 3090 GPU. I want to know that how long does it take to train your network from scratch?
Hi, team
Thanks for sharing the wonderful project!
I followed the README and successfully configured the env. However, when I was running the Database version Demo (Objetc class: pan, task: pour) by: python gcngrasp/demo_db.py --cfg_file cfg/eval/gcngrasp/gcngrasp_split_mode_o_split_idx_0.yml --obj_name pan --obj_class saucepan --task pour_, I met this BUG:
ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 1 dimensions. The detected shape was (32,) + inhomogeneous part.
################################################
Here is the content shown in my terminal when I was running the demo:
-task pour
algorithm_class: GCNGrasp
base_dir: gcngrasp/../data
batch_size: 16
dataset_class: GCNTaskGrasp
distrib_backend: dp
embedding_mode: 2
embedding_model: numberbatch
embedding_size: 300
epochs: 50
folder_dir: taskgrasp
gcn_conv_type: GCNConv
gcn_num_layers: 6
gcn_skip_mode: 0
gpus: [0]
graph_data_path: kb2_task_wn_noi
include_reverse_relations: True
instance_agnostic_mode: 1
log_dir: checkpoints/
model:
use_normal: False
use_xyz: True
name: gcngrasp_split_mode_o_split_idx_0_
num_points: 4096
optimizer:
bn_momentum: 0.5
bnm_clip: 0.01
bnm_decay: 0.5
decay_step: 20000.0
lr: 0.0001
lr_clip: 1e-05
lr_decay: 0.7
weight_decay: 0.0001
patience: 50
pc_scaling: True
pretrained_weight_file:
pretraining_mode: 0
sampling_radius: 2
split_idx: 0
split_mode: o
split_version: 1
subgraph_sampling: True
use_class_list: True
use_task1_grasps: True
weight_file: checkpoints/gcngrasp_split_mode_o_split_idx_0__2023-08-15-18-59/weights/ckpt_epoch_42.ckpt
weighted_sampling: True
########################
Namespace(cfg_file='cfg/eval/gcngrasp/gcngrasp_split_mode_o_split_idx_0.yml', data_dir='gcngrasp/../data/sample_data', obj_class='saucepan', obj_name='pan', task='pour')
Unable to find clean pc and grasps
Traceback (most recent call last):
File "gcngrasp/demo_db.py", line 273, in
main(args, cfg)
File "gcngrasp/demo_db.py", line 126, in main
pc, grasps = load_pc_and_grasps(os.path.join(data_dir, 'pcs'), obj_name)
File "gcngrasp/demo_db.py", line 97, in load_pc_and_grasps
grasps = farthest_grasps(
File "/home/zhengshen/GraspGPT_public/gcngrasp/data/../../geometry_utils.py", line 230, in farthest_grasps
grasps_fps = cluster_grasps(grasps, num_clusters=num_clusters)
File "/home/zhengshen/GraspGPT_public/gcngrasp/data/../../geometry_utils.py", line 107, in cluster_grasps
output_grasps = np.asarray(output_grasps)
ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 1 dimensions. The detected shape was (32,) + inhomogeneous part.
################################################
When I tried the other two examples (Objetc class: spatula, task: scoop, Objetc class: mug, task: drink), I also met the same BUG.
Could you please help me to solve this BUG?
Best regards,
Zhengshen
Hi,
thanks for the great work, I got the following error during evaluation 'No such object description dir: gcngrasp/../data/taskgrasp/obj_gpt_v2/scrub_brush/descriptions/5', where should I download this obj_gpt_v2 folder?
I have tried the lite version of GraspGPT, and the performance is relatively low. Therefore, I wonder when will you open source the full version of GraspGPT (i.e. the attention-based implementation of the Task-Oriented Grasp Evaluator). Thank you !
I re-implement the Task-Oriented Grasp Evaluator (TGE) module proposed in your paper, following the architecture depicted in your letter. Howerer, the performance is even worse. I am sure that I follow all the details (e.g. the dimension, the multi-head attention). I wonder if there are any more information about the TGE module, and what should I do to achieve the results you provided in the paper.
By the way, the links for downloading the LA-TaskGrasp and the checkpoints are invalid.
Hi team, thank you so much for your work.
I got a problem as the tittle shows. A lot of search work has been done, I eventially found it is an old problem of PointNet2. (see https://github.com/erikwijmans/Pointnet2_PyTorch/issues?q=furthest_point_sampling_kernel_wrapper)
I believe the problem is caused by the version of python, pytorch and CUDA.
I followed your instruction and tried:
1. Python 3.7.1 + torcu 1.11.0 + CUDA 11.3
2. Python 3.7.9 + torch 1.11.0 + CUDA 11.3
3. Python 3.7.9 + torch 1.11.0 + CUDA 10.0
but there is no lucky.
My GPU is 4090.
Please can I ask what specific version of python you used? And if you can advise me of any hint for the problem, it would be much appreciated.
Hi, thank you for open-sourcing the GraspGPT code for re-implementation. I noticed that you recently released a more advanced method, FoundationGrasp. It would be grateful if you could release its code as well.
A declarative, efficient, and flexible JavaScript library for building user interfaces.
๐ Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. ๐๐๐
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google โค๏ธ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.