Skip to content

Latest commit

 

History

History
139 lines (125 loc) · 4.17 KB

models_datasets_structure.md

File metadata and controls

139 lines (125 loc) · 4.17 KB

models and datasets structure

It's useful to know what files the repository expects in models and datasets. These can be extended as needed.

models

Models are written out to the models directory.

$ tree models
models
|-- backup
|   `-- model_dla34_cornell.pth
|-- ctdet_coco_dla_2x.pth
|-- model_alexnet_ajd.pth
|-- model_alexnet_cornell.pth
|-- model_dla34_ajd.pth
|-- model_dla34_cornell.pth
|-- model_hg104_ajd.pth
|-- model_hg104_cornell.pth
|-- model_hg52_ajd.pth
|-- model_hg52_cornell.pth
|-- model_resnet18_ajd.pth
|-- model_resnet18_cornell.pth
|-- model_resnet50_ajd.pth
|-- model_resnet50_cornell.pth
|-- model_vgg16_ajd.pth
`-- model_vgg16_cornell.pth

Note that the copy of the models in the archive has a typo for the model_dla34_cornell.pth file. This should be corrected from model_dl34_cornell.pth to model_dla34_cornell.pth.

datasets

This is the top level for the datasets directory.

$ tree datasets -L 4
datasets
|-- Cornell
|   `-- rgd_5_5_5_corner_p_full
|       `-- data
|           |-- Annotations
|           |-- ImageSets
|           `-- Images
`-- Jacquard
    `-- coco
        `-- 512_cnt_angle
            |-- test
            `-- train

We note that there are a significant number of files under the Annotations directory.

$ ls datasets/Cornell/rgd_5_5_5_corner_p_full/data/Annotations | wc -l
110625

The Jacquard dataset is a bit different, and is split across a test and train directory.

$ tree datasets/Jacquard -L 5
datasets/Jacquard
`-- coco
    `-- 512_cnt_angle
        |-- test
        |   |-- annotations
        |   |   |-- Jacquard_Dataset_0
        |   |   |-- Jacquard_Dataset_1
        |   |   |-- Jacquard_Dataset_10
        |   |   |-- Jacquard_Dataset_11
        |   |   |-- Jacquard_Dataset_2
        |   |   |-- Jacquard_Dataset_3
        |   |   |-- Jacquard_Dataset_4
        |   |   |-- Jacquard_Dataset_5
        |   |   |-- Jacquard_Dataset_6
        |   |   |-- Jacquard_Dataset_7
        |   |   |-- Jacquard_Dataset_8
        |   |   `-- Jacquard_Dataset_9
        |   |-- grasps_test2018
        |   |   |-- Jacquard_Dataset_0
        |   |   |-- Jacquard_Dataset_1
        |   |   |-- Jacquard_Dataset_10
        |   |   |-- Jacquard_Dataset_11
        |   |   |-- Jacquard_Dataset_2
        |   |   |-- Jacquard_Dataset_3
        |   |   |-- Jacquard_Dataset_4
        |   |   |-- Jacquard_Dataset_5
        |   |   |-- Jacquard_Dataset_6
        |   |   |-- Jacquard_Dataset_7
        |   |   |-- Jacquard_Dataset_8
        |   |   |-- Jacquard_Dataset_9
        |   |   `-- train_grasps_test2018_6_11.tar
        |   |-- instances_grasps_test2018.json
        |   |-- instances_grasps_test2018_edge_denseanno_filter.json
        |   `-- instances_grasps_test2018_filter.json
        `-- train
            |-- annotations
            |   |-- Jacquard_Dataset_0
            |   |-- Jacquard_Dataset_1
            |   |-- Jacquard_Dataset_10
            |   |-- Jacquard_Dataset_11
            |   |-- Jacquard_Dataset_2
            |   |-- Jacquard_Dataset_3
            |   |-- Jacquard_Dataset_4
            |   |-- Jacquard_Dataset_5
            |   |-- Jacquard_Dataset_6
            |   |-- Jacquard_Dataset_7
            |   |-- Jacquard_Dataset_8
            |   `-- Jacquard_Dataset_9
            |-- grasps_train2018
            |   |-- Jacquard_Dataset_0
            |   |-- Jacquard_Dataset_1
            |   |-- Jacquard_Dataset_2
            |   |-- Jacquard_Dataset_3
            |   |-- Jacquard_Dataset_4
            |   `-- Jacquard_Dataset_5
            |-- instances_grasps_train2018_edge_denseanno_filter.json
            `-- instances_grasps_train2018_filter.json

$ ls datasets/Jacquard/coco/512_cnt_angle/test/annotations/Jacquard_Dataset_0/ | head -n10
1a0312faac503f7dc2c1a442b53fa053
1a0710af081df737c50a037462bade42
1a2a5a06ce083786581bb5a25b17bed6
1a30adabf5a2bb848af30108ea9ccb6c
1a3efcaaf8db9957a010c31b9816f48b
1a46011ef7d2230785b479b317175b55
1a477f7b2c1799e1b728e6e715c3f8cf
1a4daa4904bb4a0949684e7f0bb99f9c
1a5327b328cd97d084c3569473be6c23
1a5f561ce4cbca2625c70fb1df3f879b

Each dataset in the training set represents roughly 550 images, while each dataset in the test set represents roughly 900 images.