search

ProjectSidewalk / sidewalk-cv-assets19

Repo for our ASSETS'19 paper applying ResNet to Project Sidewalk data
5 stars 6 forks source link

.json file produced by bulk_extract_crops() #36

Open chrissuda opened 4 years ago

chrissuda commented 4 years ago

The following .json file is produced by running bulk_extract_crops(): {"crop size": 718.0457792111467, "sv_x": 4307.692307692308, "sv_y": -369.2307692307692, "pano id": "_ahZ5V6MINus1_n1Ve7Hrw", "pano yaw": 123.37543869018549, "crop_y": 4465.2307692307695, "crop_x": 9922.645606303417}

As you can see, some keys are missing compared with yours: {"dist to cbd": 4.094305012075221, "bearing to cbd": 64.74029765051874, "crop size": 1492.1348109969322, "sv_x": 9300.0, "sv_y": -1500.0, "longitude": -76.967779, "pano id": "__1c3_5IArbrml1--v7meQ", "dist to intersection": 23.45792342820621, "block middleness": 42.95327159437733, "latitude": 38.872448, "pano yaw": -179.67633056640602, "crop_y": 4828.0, "crop_x": 2655.9685763888974}

How can I solve this problem? I have tried it on both DC and Seattle datasets. Both yield only 7 keys in .json files And the strange things were: I can test your model by feeding the incomplete data(.json ).

galenweld commented 4 years ago

Hmm. What is the bulk_extract_crops function you're using?

Either way, the additional keys you're missing are the "geographic and positional features" described in the paper. I would recommend not using them or the versions of the models that require them, these extra features add considerable complexity to the model training and usage pipeline, and we found that they have very little impact on performance.

On Wed, Sep 9, 2020 at 6:51 PM chrissuda notifications@github.com wrote:

The following .json file is produced by running bulk_extract_crops(): {"crop size": 718.0457792111467, "sv_x": 4307.692307692308, "sv_y": -369.2307692307692, "pano id": "_ahZ5V6MINus1_n1Ve7Hrw", "pano yaw": 123.37543869018549, "crop_y": 4465.2307692307695, "crop_x": 9922.645606303417}

As you can see, some keys are missing compared with yours: {"dist to cbd": 4.094305012075221, "bearing to cbd": 64.74029765051874, "crop size": 1492.1348109969322, "sv_x": 9300.0, "sv_y": -1500.0, "longitude": -76.967779, "pano id": "__1c3_5IArbrml1--v7meQ", "dist to intersection": 23.45792342820621, "block middleness": 42.95327159437733, "latitude": 38.872448, "pano yaw": -179.67633056640602, "crop_y": 4828.0, "crop_x": 2655.9685763888974}

How can I solve this problem? I have tried it on both DC and Seattle datasets. Both yield only 7 keys in .json files And the strange things were: I can test your model by feeding the incomplete data(.json ).

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/ProjectSidewalk/sidewalk-cv-assets19/issues/36, or unsubscribe https://github.com/notifications/unsubscribe-auth/AACXTSRGYSECPRAAZNUBTZDSFAWKDANCNFSM4RD2UPXQ .

chrissuda commented 4 years ago

bulk_extract_crops() is a cropper function stated in README.md in this repository under "SetUp" section: These crops and .json files can be produced easily and simultaneously using the bulk_extract_crops function from GSVutils/utils.py. This function takes as input a .csv with the following columns:

BTW, what did "sv_x sv_y" and "crop_x crop_y" in the .josn file refer to? And what is the difference between them (sv, crop)?

Appreciated your explanation.

galenweld commented 3 years ago

Apologies for the confusion. There are a number of different coordinate systems in use, which gets a bit messy. The SV (short for street view) coordinates are the ones used by google, which if I recall correctly place the origin at the horizon, facing north.

The crop coordinates (if I recall correctly) I believe transform the sv coordinates into pixel coordinates on the image which can then be used to extract a crop. Let me know if that doesn't line up with what you're seeing.