Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Trained model #3

Open
WeedsInML opened this issue Aug 24, 2022 · 7 comments
Open

Trained model #3

WeedsInML opened this issue Aug 24, 2022 · 7 comments

Comments

@WeedsInML
Copy link

Hello.
This is very nice work.
Could you please provide your trained model?

@zerchen
Copy link
Owner

zerchen commented Nov 23, 2022

Hi,

Yes, I will upload my trained model soon. You could first give a try to use my preprocessed data (https://drive.google.com/drive/folders/1GoaA6vB6TwAAHmaobVo5GjRoCq2wT21R) to do the training.

Best,
Zerui

@EAST-J
Copy link

EAST-J commented Mar 9, 2023

Hi,

Yes, I will upload my trained model soon. You could first give a try to use my preprocessed data (https://drive.google.com/drive/folders/1GoaA6vB6TwAAHmaobVo5GjRoCq2wT21R) to do the training.

Best, Zerui

Hi,
Have you uploaded the trained model?

@zerchen
Copy link
Owner

zerchen commented Mar 9, 2023

Hi,

Thanks for your reminder. I have put the trained model on ObMan on here. Please give a note if it does not work or not compatible with this codebase. The checkpoints trained on dexycb will be released with a new codebase, which is more user-friendly.

Best,
Zerui

@RNGrunshen
Copy link

Hi,

Thanks for your reminder. I have put the trained model on ObMan on here. Please give a note if it does not work or not compatible with this codebase. The checkpoints trained on dexycb will be released with a new codebase, which is more user-friendly.

Best, Zerui

hello how can I use the trained model to evaluate

@happyiminjay1
Copy link

happyiminjay1 commented Jun 1, 2023

Hi,

Thanks for your reminder. I have put the trained model on ObMan on here. Please give a note if it does not work or not compatible with this codebase. The checkpoints trained on dexycb will be released with a new codebase, which is more user-friendly.

Best, Zerui

Hello,

I think the specs.json file which is uploaded on the experiments/obman/*.json file is not compatible to the trained model.

Can you upload the right specs.json file?

Thanks.

@zerchen
Copy link
Owner

zerchen commented Jun 1, 2023

Hi,

Thanks for your interests. Actually, I have not touched this codebase for a while. Some configurations might have changed during iterations of different versions.
I suggest using the latest codebase, which provides a cleaned implementation of AlignSDF. To test the model, you could download the directory and place it under $gSDF_ROOT/outputs. Then, please follow the instructions written here to test and evaluate the model. Please note that the current checkpoint is only trained with 750 epochs, instead of 1600 epochs used in the original paper. More checkpoints will come later.

Best,
Zerui

@happyiminjay1
Copy link

Thank you for the clear explanation! I will try on the latest codebase.

Someone who needs the specs.json file can use the following text.

{
"Description": "3D hand reconstruction on the mini obman dataset.",
"DataSource": "data",
"ImageSource": "rgb",
"TrainSplit": "experiments/splits/obman_80k.json",
"Dataset": "obman",
"ModelType": "1encoder2decoder",
"LMDB": true,
"ImageSize": [256, 256],
"SdfScaleFactor": 7.018621123357809,
"LatentSize": 256,
"PointFeatSize": 6,
"EncodeStyle": "nerf",
"ScaleAug": false,
"PoseFeatSize": 15,
"SnapshotFrequency": 500,
"LogFrequency": 5,
"LogFrequencyStep": 10,
"NumEpochs": 1600,
"Backbone": "resnet18",
"Resume": "latest.pth",
"Freeze": "none",
"PixelAlign": false,
"AdditionalSnapshots": [
100,
500
],
"AdditionalLossStart": 1201,
"SamplesPerScene": 2000,
"ScenesPerBatch": 64,
"DataLoaderThreads": 5,
"ClampingDistance": 0.05,
"HandBranch": true,
"ObjectBranch": true,
"ObjectPoseBranch": true,
"ManoBranch": true,
"DepthBranch": false,
"Render": false,
"ClassifierBranch": false,
"ClassifierWeight": 0.005,
"PenetrationLoss": false,
"ContactLoss": false,
"IndependentObjScale": false,
"IgnorePointFromOtherMesh": false,
"HandSdfWeight":0.5,
"ObjSdfWeight":0.5,
"JointWeight":0.5,
"VertWeight":0,
"ShapeRegWeight":0.0000005,
"PoseRegWeight":0.00005,
"SegWeight":0.004,
"ObjCenterWeight":1,
"ObjCornerWeight":0,
"PenetrationLossWeight":15.0,
"ContactLossWeight":0.005,
"DisableAug":false,
"BackgroundAug":false,
"NetworkSpecs": {
"dims": [
512,
512,
512,
512
],
"dropout": [
0,
1,
2,
3
],
"dropout_prob": 0.2,
"norm_layers": [
0,
1,
2,
3
],
"latent_in": [
2
],
"num_class": 6,
"xyz_in_all": false,
"use_tanh": false,
"latent_dropout": false,
"weight_norm": true
},
"LearningRateSchedule": [
{
"Type": "Step",
"Initial": 0.0001,
"Interval": 600,
"Factor": 0.5
}
]
}

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

5 participants