Tools Recomendation ???

[HappySamuel]

What kind of tools are recommended for preparing the .json files in sample_data/ directory relevant to the .png / .jpg images before proceed to next stage of using train.py ?

I tried the labelme, but the .json file generated looks like below, it isn't same as the one show in the sample_data/ directory ?

{
"version": "3.16.7",
"flags": {},
"shapes": [
{
"label": "flat_screwdriver",
"line_color": null,
"fill_color": null,
"points": [
[
2805.8688524590166,
678.9508196721312
],
[
2810.786885245902,
756.0
],
[
2800.9508196721313,
816.655737704918
],
[
2796.032786885246,
1315.016393442623
],
[
2835.377049180328,
1315.016393442623
],
[
2884.5573770491806,
1365.8360655737706
],
[
2856.688524590164,
1451.0819672131147
],
[
2858.3278688524592,
1510.0983606557377
],
[
2873.0819672131147,
1587.1475409836066
],
[
2892.754098360656,
1693.704918032787
],
[
2891.1147540983607,
1964.1967213114754
],
[
2843.5737704918033,
2011.7377049180327
],
[
2845.2131147540986,
2034.688524590164
],
[
2805.8688524590166,
2039.6065573770493
],
[
2746.8524590163934,
2034.688524590164
],
[
2710.786885245902,
2011.7377049180327
],
[
2715.7049180327867,
1996.983606557377
],
[
2681.27868852459,
1947.8032786885246
],
[
2691.1147540983607,
1674.032786885246
],
[
2714.065573770492,
1583.8688524590164
],
[
2727.1803278688526,
1518.295081967213
],
[
2723.9016393442625,
1452.72131147541
],
[
2696.032786885246,
1377.311475409836
],
[
2735.377049180328,
1319.9344262295083
],
[
2773.0819672131147,
1311.7377049180327
],
[
2771.44262295082,
818.2950819672132
],
[
2758.3278688524592,
762.5573770491803
],
[
2768.1639344262294,
680.5901639344263
]
],
"shape_type": "polygon",
"flags": {}
}
],
"lineColor": [
0,
255,
0,
128
],
"fillColor": [
255,
0,
0,
128
],
"imagePath": "N1_0.jpg",
"imageHeight": 2848,
"imageWidth": 4272
}

[TontonTremblay]

In the repo we have two solutions, one with blender and the other one with nvisii.

[HappySamuel]

Hi @TontonTremblay

Thanks for the info. I have toilet bowl images, if i want to detect it and get the pose of it when train with DOPE, which script of nvisii shall i use to prepare the dataset .json file ?

[TontonTremblay]

Do you have a 3d model?

…

On Tue, Jan 7, 2025 at 17:56 HappySamuel ***@***.***> wrote: Hi @TontonTremblay <https://github.com/TontonTremblay> Thanks for the info. I have toilet bowl images, if i want to detect it and get the pose of it when train with DOPE, which script of nvisii shall i use to prepare the dataset .json file ? — Reply to this email directly, view it on GitHub <#396 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABK6JIHCBAXUK6I43YCDOBT2JSATVAVCNFSM6AAAAABUXCNNZCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDKNZWGU2TEMRVGQ> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[HappySamuel]

Do you have a 3d model?

Nope, i don't have. Currently only have .jpg files. Do i need 3d model to do use with DOPE ? I thought only isaac_ros_foundationpose will need 3d model.

[TontonTremblay]

our tools assume a 3d models. So without one I dont think I can help you. sorry.

[HappySamuel]

OK, then i try to generate one. What are the file extensions for the 3d models shall i prepare ? Must it be exact the same as per images ? Or can it be different ?

[HappySamuel]

Hi @TontonTremblay

I want to ask, if the model trained by dope, can it be use for centerpose as well ? Or any modifications need to be amend ?

[TontonTremblay]

I think both models are trained on the same data structure.

[TontonTremblay]

As for the file, obj should be ok.