Open chaudatascience opened 2 weeks ago
Hi, Just checking in to see if any updates on this.
I have trained a model, and got the output at my_log_dir
.
However, I can't generate the 3D objects. Can you provide a code to generate it, such as in URDF?
I've tried using obj_retrieval.py
as mentioned in issue #1, but still couldn't get the result.
Some issues: How should I set the values for requirement_dict
, dataset_dir
, and hashbook_path
? Also, what should the value of gt_file_name
be in case it is unavailable (i.e., generating new objects)?
Here's what I've tried so far:
It'd be great if you could provide a runnable example for it.
Thank you!
Chau
requirement_dict_path = "my_log_dir/run1/cage/v1/images/predict/cond_graph/10000/Dishwasher/06fee7d302703f9fbdebad50b50439e1.json"
requirement_dict = json.load(open(requirement_dict_path, "r"))
dataset_dir = "my_log_dir/run1/cage/v1/images/predict/cond_graph/10000/"
hashbook_path = "cage/indexes/hash_table.json"
num_states = 5
metric_transform_plucker = False
metric_compare_handles = False
metric_iou_include_base = False
metric_scale_factor = 10
metric_num_samples = 10000
keep_top = 5
gt_file_name = "train_renumber.json"
verbose = True
obj_candidates = find_obj_candidates(
requirement_dict,
dataset_dir,
hashbook_path,
num_states,
metric_transform_plucker,
metric_compare_handles,
metric_iou_include_base,
metric_scale_factor,
metric_num_samples,
keep_top,
gt_file_name,
verbose,
)
parts_to_render = pick_and_rescale_parts(
requirement_dict, obj_candidates, dataset_dir, gt_file_name, verbose
)
print(parts_to_render)
Hi Chau! Thanks for your interest in our work and for suggesting the demo code. We are working on it now. Will let you know once they are released.
Dear Jiayi,
Thanks for your response! Looking forward to the new update.
Best, Chau
Hi, Thank you and congrats on the great work!
Can you provide the checkpoint of the model and the code to generate the objects?
Thank you! Best, Chau