- Working vss init with some refactored library calls
(
3730fcb)
- Add --use-cleanvision as option as it is not appropriate for all datasets
(
eb97d9a)
- Handle empty data versions
(
2527f4a)
- uav: Reduce conf to improve detection
(
36c1c5e)
- uav: Handle models hosted locally instead of huggingface
(
43ed3cd)
- uav: Sam based segmentation to fix too large boxes on some localizations
(
69e03f0)
- Handle processing non mission collections for evaluating models
(
5fa21c2)
- uav: Handle bogus path to detections or clusters and use version defined in config.yml
(
cc4cba0)
- uav: Support either gpu/cpu processing and load model from local mount for speedup
(
82da909)
- Correct clean imagesearch
(
321cbe5)
- i2map: 200 meter cluster bulk and load pipeline; processes and loads top two scores for
further exploration of fish clustering and improved training data
(
140cf77)
- i2map: Sdcat switch to 8 block model and skip blurry
(
76c5b33)
- i2map: Sdcat switch to 8 block model and skip blurry
(
fa95645)
- Remove 180 rotation in augmentation; this is handled in training now
(
507bacd)
- uav: Switch to 8 block size model /mnt/DeepSea-AI/models/UAV/mbari-uav-vit-b-8-20241231
(
490a345)
- i2map: Download filtered results by vars-labelbot and group
(
11a97bc)
-
Generates csv of stats to share with team;combine stats.json in nested directories of multiple versions (
50931d2) -
Support --version override in download_crop_pipeline.py (
a16c3ee) -
uav: Filter results by min score before loading (
950ea3d)
- i2map: Remove dark images
(
9a7c299)
- Updated with latest justfile
(
bf649c4)
- uav: Switch to latest mbari-uav-vit-b-16-20241230 model
(
948ba99)
-
Add stats generation after download in download_crop_pipeline.py (
86237d3) -
Clean only versioned dir not parent (
5b830e1) -
Remove voc crop and replace with download in download_crop_pipeline.py (
016d515) -
bio: Bound boxes before redis load (
fcf3678)
- uav: Remove near duplicates to reduce training time
(
302479d)
- Added cleanvision to all projects
(
ca24a03)
- Add crop resize from aidata to download crop pipeline
(
188044e)
- cfe: Rachelcarson/2023/07 ISIIS video load
(
ccdf54e)
- cfe: Correct platform parse based on directory naming convention
(
5221134)
- cfe: Added common args for cfe and added missing mission_name
(
5aff45d)
- cfe: Video loading pipeline initial check-in
(
a7200ab)
-
Handle failed video capture and simplify args (
14dd788) -
Switch to frame stride and correct conversion of tracking coordinates (
485cbfb)
- Fix build conflict in apache-beam/transformers/pyyaml
(
01b19df)
- bio: Parallel crop
(
764f524)
- Remove short low scoring tracks
(
c8aed2d)
- Move crop to GPU
(
bd11c3e)
- bio: Significant refactor of video processing pipeline into more modular design with callbacks
and moved most data to GPU where possible for speed-up.
(
1b3d972)
- bio: Handle both fractional and non-fractional seconds from metadata query and bump empty
frames to 5
(
4c4dbf0)
- bio: Updated just recipe to include install and processing of single videos and dives
(
6540634)
- bio: Pass through separate score threshold for det/track
(
2d34434)
- bio: Some refactoring, plus added second label_s/score_s based on gcam which is proving to be
informative and remove any detections that are blurry which are not useful for detection/tracking
(
f2799b8)
- Added vits models to project where model exists and clean for sdcat pipeline
(
0a16c0a)
- uav: Jelly heat map
(
bebb389)
- uav: Skip over exemplars when loading clustered detections
(
ee81689)
- uav: Increase min saliency to 300 in detect
(
e622e20)
- bio: Add vits model to strided video track
(
4c9027a)
- uav: Switch to one class detection model and --use-vits in clustering
MBARI-org/mbari-uav-vit-b-16
(
80da5c2)
- Correct parsing of download args
(
09f0a90)
- Support vss load all
(
3d748aa)
- uav: Bump min_std from 2. to 4. to reduce detections and run vss after cluster for auto label
(
e735f94)
- Remove is_near_duplicate_issue from clean and add 180 deg augmentation
(
72253e5)
- Correct check for exemplar files with v0.38.2 perf improvement
(
86f585a)
- More cleanining of recipes and updated dep build to include biotrack
(
8f3b2e7)
- Add missing dependency
(
c118d35)
- Updated recipes
(
a0bb70c)
- bio: Added two-stage strided track pipeline
(
1a9cc1b)
- Skip cluster in vss pipeline for < 2000 detections
(
2591116)
- Add default EXIF fields if missing
(
580643e)
- Add originating bounding box to crop for use in downstream processing in EXIF comment field, e.g.
UserComment: b'bbox:534,16,817,300'
(
ad49140)
- Correct dictionary item for vss project
(
afd99e0)
- Added vss to UAV processing pipeline
(
df81cc1)
-
Mostly some cleaning of args and logging for vss_predict_pipeline.py but also bumped the batch size to 20 for faster prediction (
84608ad) -
Removed color jitter from aug and remove all issues, dark, clurry, duplicate, near-duplicate and only cluster is > 500 examples for vss (
7a38228)
- Correct args to download_crop_pipeline.py
(
c6a4fe7)
- Correct check for parent of bound volumes and pass through args to download_crop_pipeline.py
(
9906488)
- Pass through additional args in download_pipeline.py
(
ed7ae48)
- bio: Correct max_seconds args
(
03dd84e)
- bio: Correct normalization of localization and handle exceptions in vss
(
3349391)
-
Added co-tracker submodule (
4bcd71a) -
Added submodule for co-tracker and better formatting for cluster-uav (
8aaef9b)
- bio: Added support for --skip-load and --max-seconds to save just localizations
(
c5f6cda)
- uav: Remove any low saliency detections from clustering
(
43b316c)
- Clean vss images with cleanvision defaults removing all dark and blurry images
(
1ccb994)
- uav: Exclude unused classes, lower threshold for vss to allow more rare detections, and
cluster everything
(
c7d12b6)
- bio: Removed unused args
(
1038aa7)
- bio: Pass through addtional args
(
3bf1b74)
- Minor logging correction
(
544f3b7)
- bio: Added --allowed-classes animal with --class-remap "{'animal': 'marine organism'}"
(
1306116)
- Load all detections if less than 100 total examples
(
7ec6053)
-
cfe: Delete all media by depth (
41a60f9) -
i2map: Change group MERGE_CLASSIFY to NMS (
6fb7265) -
uav: Delete all loc in media (
6323700)
- Updated README with latest recipes
(
5eda62d)
- Added vss remove e.g. j remove-vss uav --doc 'doc:boat:*'
(
06b02d9)
- Add --min-variance to vss-init
(
c838aae)
- bio: Correct vignette logic
(
e18ff74)
- bio: Normalize coords
(
b42cec4)
- bio: Add --remove-vignette to remove vignette detections
(
675a501)
- bio: Add --skip-vss to skip over second stage
(
809188f)
- cfe: Load media+boxes of mined rare class labels
(
c8b9824)
- bio: Add blurriness removal
(
113e7f8)
- bio: Crop pass to vss, and pass confidence detections less than .9
(
69c4b99)
- cfe: Switch from video to frame mining on images with depth
(
bd5ee13)
- bio: Simple script to delete a particular version
(
39b0788)
- Continue to load exemplar when missing data as some labels may be missing
(
e3fbdbd)
- Added load vss only pipeline
(
23c9b78)
- Added vss optimizer
(
e3dc508)
- Handle zero cost saliency
(
2d990e4)
- Remove low saliency matches from downloaded data to support external data integration
(
7cb5440)
- bio: Set saliency score to 1000 for download
(
b86f118)
- cfe: Add download_isiis_label_metadata.py for Fernanda
(
6ee109c)
- Handle alternative image directory for voc crops
(
0777493)
- Handle alternative image directory downloads and skipping download for VSS initialization
(
4f74e68)
- cfe: Switch to depth labeled images for data mining
(
9dd60ae)
- Minor typo fix
(
d31b650)
- bio: Added download for top concepts
(
3f0372b)
- Correct search pattern completion
(
c8d236c)
- Added support for optional database update with --update to saliency pipeline
(
ec2c78a)
-
Added download only pipeline (
3d85098) -
Support search pattern left off of saliency pipeline (
4bb9685)
- Added support for updating the saliency attribute for any project with variable block size, std,
and rescaling; run with just compute-saliency uav --scale-percent 50 --min-std 2.0 --block-size 39
--voc-search-pattern <path to your voc/*.xml>
(
6bcb0e4)
- bio: Pass config dict
(
6dcf8fe)
-
bio: Correct clean-up (
7c2684c) -
bio: Correct frame number which was lost during perf update (
0531405)
- bio: Clean-up frame grabs and only run on < 200 meters
(
729f337)
- bio: Switch to ffmpeg generation of highest quality jpeg ffmpeg for speed-up
(
42c23cc)
- bio: Check depth before starting
(
d9d74eb)
- bio: Minor correction to the endpoint
(
f22beeb)
- Added saliency computation and updated docs
(
1920227)
- Support pass through of environment variables to docker
(
19de539)
- bio: Added missing TATOR_TOKEN
(
21fcd31)
-
bio: Add test recipe for processing single video with two-stage pipeline (
41ce64f) -
bio: Resolve video to path (
7d800fc)
- bio: Added ancillary data, slightly faster video seek and better logging for multiproc
(
28e0772)
- bio: Updated README with command line changes
(
1aa0b69)
- bio: More conservative threshold for second stage vss
(
d8be83f)
- bio: Added expc metadata docker build and test data
(
8d49c73)
- bio: Added missing yaml entry and fix typo
(
811b988)
- uav: Final assigment for UAV version saliency_MBARI/uav-yolov5-30k_VSS
(
5a8b2e8)
- uav: Load all detections in single docker instance
(
8d3c7c3)
- bio: Added vss second stage
(
c5defea)
- bio: Added support to grab data version by name
(
10a5c09)
- bio: Fixed refactoring bugs
(
9f85436)
- cfe: Reduce detections with saliency, min/max area adjustments
(
676a5e9)
- uav: Add Kelp,Bird pass through
(
0c152eb)
- uav: More memory efficient multiproc crop
(
eab6f5d)
- uav: Faster vss with multiproc crop, correct handling of csv output and pipeline conditional
(
4b6f4f1)
- Correct best score average
(
2dff7f3)
- Working two-stage vss pipeline detect for UAV
(
d9c288e)
- Reduce min_std to 2.0 to find more elusive detections like batrays
(
45777ff)
- cfe: Replace depth by name in CFE images
(
d41cba0)
- uav: Added vss to predict pipeline
(
ca52762)
- Add extra args to predict vss and correct download crop config
(
4d0864b)
- Correct termination
(
08e0f5e)
- Added isiis mine script
(
3105d01)
- Correct config for i2map host
(
bd0a2be)
- Correct threshold for scoring vss
(
a7cb0db)
-
Correct path for config setup (
acf25a2) -
Pass through model name and skip if no labels for graceful exit (
ffe5e1c)
- Add metadata fix for UAV images, correct path for sdcat config and updated missions2process.txt
(
fc42612)
- Added support for more config checks and some refactoring
(
300b46f)
- Added support for block size 39 and std of 5.0 for uav
(
6e7fbe6)
- Export confused predictions and comments to a csv for further processing
(
9457a3b)
- Flag missing exemplar file, break after load and other refactoring
(
7acc573)
-
Add console log to vss accuracy, get class names from redis not config, and handle empty predictions (
cc5a517) -
Add more checks for required keys in config (
4446d1d) -
Add override for download crop pipeline for various arguments (
7a307db) -
Plot names in centroids of each class (
8c5dd39) -
Working download_crop_pipeline.py with updates for voc-cropper:0.4.3 (
76ee970)
-
Better handling of missing classes and bump low exemplars to 10 (
f486fdd) -
Report correct number augmented labels and change from 1 to 2 batch clusters (
a990b70)
-
Add augment data clean in case not run before metric (
a2dcbea) -
Added the guano for eval (
c0c9383) -
Correct redis port for cfe project (
c4d2e3f)
-
Add config for bulk i2map (
8df2e17) -
Added skip to download and crop pipeline and add remove data defaults - these should be specified by project (
dc29285)
- Removed any augmented data for init vss and other minor improvements to plot names
(
2f040d0)
-
Added new uav missions and fixed mission parse for loading images (
940ed36) -
Correct config ath (
1c53abb) -
Correct uav imports after refactoring munged and merged aidata config (
9a8bf29)
- Added simclr_augmentations like augmentation and other minor formatting changes
(
0d4493d)
- Added missing pip install to install recipe
(
e4bf1d2)
- Added confusion matrix and more cleaning of metrics code
(
d205072)
- Minor typo fix
(
9e7a728)
- Add retry to crop with 30 second delay
(
193542b)
- Added missing files and reduce docker log names to daily
(
1847fee)
- Added example tsne plot
(
779e9ae)
- Initial commit
(
90e809d)