opengvlab / hulk Goto Github PK

Thanks for your work, it was an error on my side I accidentally deleted decoder: peddet_cfgs

It seems that there is a bug here.

Hulk/core/solvers/utils/pos_tester_dev.py", line 134, in evaluate
image_id = self.dataset.name2id[image_paths[i][len(self.dataset.img_prefix):]]

According line 53 self.dataset = config.evaluation.cfg.dataset, the initialization of self.dataset is based on the test YAML file, but no corresponding configuration was found in the YAML file.

Could you demonstrate how to initialize model by pythonic way and get predictions on images tensor for specific task?
Best!

More specifically, I want to finetune Hulk for human parsing on my custom dataset. What should I do?

The paper mentions that Backbone uses the same ViT-base as UniHCP, but when looking at the code, generate_random_masks is used in vit to shuffle. I would like to ask what the purpose of this is and if there are any more literature on this approach.

It seems that the open source weights have one ckpt per task. Is it possible to merge the weights of different tasks and only load one merged model during inference? Has this been used in practice?

Dear Authors,

First of all, thank you for sharing your code! I am trying to use it for inference, but I get an error here:

The code tries to load ./checked_par_lip_name.pth file for LIP evaluation (this is the task for which I am trying to use your model in inference mode on my own images) but this file is not part of the repository. Could you help me out here how to get this file?

Thank you for your great work.
I have a question about practical inference. Is it necessary to infer twice for keypoint, first by detection to crop the human body and then to infer the key points?

Traceback (most recent call last):
File "/mnt/data/yuxiong/Hulk/experiments/release/../..//test_mae.py", line 9, in
from core.testers import tester_entry
File "/mnt/data/yuxiong/Hulk/core/testers/init.py", line 1, in
from ..solvers.solver_mae_devnew import TesterMAEDev
File "/mnt/data/yuxiong/Hulk/core/solvers/init.py", line 3, in
from .solver_mae_devnew import SolverMAEDev
File "/mnt/data/yuxiong/Hulk/core/solvers/solver_mae_devnew.py", line 40, in
from core.solvers.utils.peddet_tester_dev import PedDetMAEEvaluator
File "/mnt/data/yuxiong/Hulk/core/solvers/utils/peddet_tester_dev.py", line 28, in
from petrelbox.io import PetrelHelper
ModuleNotFoundError: No module named 'petrelbox'

pip/conda install petrelbox doesn't work, I was trying to do the evaluation, pls provide an alternative way

if i use you model,how should i orginize the dataset

Is there any good way to train or evaluate Hulk without using Slurm?