GroundedSAM labels

[csv610]

Hello,

When I look at the labels generated by GroundedSAM, I see a huge list of floating number not the data bounding boxes in yolo format. How should I interpret the output? Is it possible to store the output inn the Yolo format?

[capjamesg]

Can you share an example of the output?

[shersoni610]

Hi, This is the label for an image containing a single object (sharp edges).

0 0.41200 0.09202 0.41100 0.09313 0.41100 0.09424 0.40800 0.09756 0.40800 0.09978 0.40700 0.10089 0.40700 0.10200 0.40600 0.10310 0.40600 0.10421 0.40500 0.10532 0.40500 0.10643 0.40400 0.10754 0.40400 0.10865 0.40300 0.10976 0.40300 0.11419 0.40200 0.11530 0.40200 0.11641 0.40000 0.11863 0.40000 0.12084 0.39900 0.12195 0.39900 0.12417 0.39800 0.12528 0.39800 0.12639 0.39700 0.12749 0.39700 0.12860 0.39600 0.12971 0.39600 0.13304 0.39500 0.13415 0.39500 0.13858 0.39600 0.13969 0.39500 0.14080 0.39500 0.15965 0.39400 0.16075 0.39400 0.17627 0.39300 0.17738 0.39100 0.17517 0.39100 0.17295 0.39200 0.17184 0.39200 0.17073 0.39100 0.16962 0.39100 0.16297 0.39000 0.16186 0.39000 0.16075 0.38800 0.16075 0.38700 0.15965 0.38600 0.15965 0.38500 0.15854 0.37900 0.15854 0.37800 0.15965 0.37300 0.15965 0.37200 0.16075 0.36700 0.16075 0.36600 0.16186 0.36300 0.16186 0.36200 0.16297 0.35700 0.16297 0.35600 0.16408 0.35100 0.16408 0.35000 0.16519 0.34800 0.16519 0.34700 0.16630 0.34500 0.16630 0.34400 0.16741 0.34000 0.16741 0.33900 0.16851 0.33600 0.16851 0.33500 0.16962 0.33300 0.16962 0.33200 0.17073 0.33000 0.17073 0.32900 0.17184 0.32500 0.17184 0.32400 0.17295 0.32100 0.17295 0.32000 0.17406 0.31900 0.17406 0.31800 0.17517 0.31600 0.17517 0.31500 0.17627 0.31200 0.17627 0.31100 0.17738 0.31000 0.17738 0.30900 0.17849 0.30700 0.17849 0.30600 0.17960 0.30400 0.17960 0.30300 0.18071 0.30100 0.18071 0.30000 0.18182 0.29800 0.18182 0.29700 0.18293 0.29600 0.18293 0.29500 0.18404 0.29300 0.18404 0.29200 0.18514 0.28900 0.18514 0.28800 0.18625 0.28700 0.18625 0.28600 0.18736 0.28500 0.18736 0.28400 0.18847 0.28200 0.18847 0.28100 0.18958 0.27700 0.18958 0.27600 0.19069 0.27500 0.19069 0.27400 0.19180 0.27300 0.19180 0.27200 0.19290 0.27100 0.19290 0.27000 0.19401 0.26800 0.19401 0.26700 0.19512 0.26600 0.19512 0.26500 0.19623 0.26400 0.19623 0.26300 0.19734 0.26000 0.19734 0.25900 0.19845 0.25700 0.19845 0.25600 0.19956 0.25500 0.19956 0.25400 0.20067 0.25300 0.20067 0.25200 0.20177 0.25000 0.20177 0.24900 0.20288 0.24600 0.20288 0.24500 0.20399 0.24300 0.20399 0.24100 0.20621 0.24000 0.20621 0.23900 0.20732 0.23600 0.20732 0.23500 0.20843 0.23400 0.20843 0.23300 0.20953 0.23200 0.20953 0.23100 0.21064 0.22900 0.21064 0.22800 0.21175 0.22600 0.21175 0.22500 0.21286 0.22400 0.21286 0.22300 0.21397 0.22200 0.21397 0.22100 0.21508 0.21800 0.21508 0.21700 0.21619 0.21600 0.21619 0.21500 0.21729 0.21300 0.21729 0.21200 0.21840 0.21100 0.21840 0.21000 0.21951 0.20900 0.21951 0.20800 0.22062 0.20600 0.22062 0.20500 0.22173 0.20400 0.22173 0.20300 0.22284 0.20200 0.22284 0.20100 0.22395 0.19900 0.22395 0.19800 0.22506 0.19600 0.22506 0.19400 0.22727 0.19300 0.22727 0.19200 0.22838 0.19000 0.22838 0.18900 0.22949 0.18800 0.22949 0.18600 0.23171 0.18500 0.23171 0.18400 0.23282 0.18200 0.23282 0.18100 0.23392 0.18000 0.23392 0.17800 0.23614 0.17700 0.23614 0.17600 0.23725 0.17500 0.23725 0.17400 0.23836 0.17300 0.23836 0.17100 0.24058 0.17000 0.24058 0.16900 0.24169 0.16800 0.24169 0.16600 0.24390 0.16500 0.24390 0.16300 0.24612 0.16200 0.24612 0.16000 0.24834 0.15900 0.24834 0.15700 0.25055 0.15600 0.25055 0.15400 0.25277 0.15300 0.25277 0.15100 0.25499 0.15000 0.25499 0.14600 0.25942 0.14500 0.25942 0.13800 0.26718 0.13700 0.26718 0.13500 0.26940 0.13400 0.26940 0.11500 0.29047 0.11500 0.29157 0.11200 0.29490 0.11200 0.29601 0.10900 0.29933 0.10900 0.30044 0.10300 0.30710 0.10300 0.30820 0.10100 0.31042 0.10100 0.31153 0.10000 0.31264 0.10000 0.31375 0.09700 0.31707 0.09700 0.31818 0.09500 0.32040 0.09500 0.32151 0.09300 0.32373 0.09300 0.32483 0.09200 0.32594 0.09200 0.32705 0.09100 0.32816 0.09100 0.32927 0.09000 0.33038 0.09000 0.33149 0.08900 0.33259 0.08900 0.33370 0.08700 0.33592 0.08700 0.33703 0.08600 0.33814 0.08600 0.33925 0.08500 0.34035 0.08500 0.34257 0.08300 0.34479 0.08300 0.34590 0.08200 0.34701 0.08200 0.34812 0.08100 0.34922 0.08100 0.35033 0.08000 0.35144 0.08000 0.35255 0.07900 0.35366 0.07900 0.35477 0.07800 0.35588 0.07800 0.35698 0.07700 0.35809 0.07700 0.35920 0.07600 0.36031 0.07600 0.36142 0.07500 0.36253 0.07500 0.36475 0.07400 0.36585 0.07400 0.36696 0.07300 0.36807 0.07300 0.36918 0.07200 0.37029 0.07200 0.37140 0.07100 0.37251 0.07100 0.37361 0.07000 0.37472 0.07000 0.37805 0.06900 0.37916 0.06900 0.38027 0.06800 0.38137 0.06800 0.38359 0.06700 0.38470 0.06700 0.38581 0.06600 0.38692 0.06600 0.38914 0.06500 0.39024 0.06500 0.39246 0.06400 0.39357 0.06400 0.39579 0.06300 0.39690 0.06300 0.39911 0.06200 0.40022 0.06200 0.40244 0.06100 0.40355 0.06100 0.40576 0.06000 0.40687 0.06000 0.40798 0.05900 0.40909 0.05900 0.41463 0.05800 0.41574 0.05800 0.42018 0.05700 0.42129 0.05700 0.42572 0.05600 0.42683 0.05600 0.43348 0.05500 0.43459 0.05500 0.43792 0.05400 0.43902 0.05400 0.45011 0.05300 0.45122 0.05300 0.45898 0.05200 0.46009 0.05200 0.47118 0.05100 0.47228 0.05100 0.48448 0.05000 0.48559 0.05000 0.50222 0.05100 0.50333 0.05100 0.52439 0.05200 0.52550 0.05200 0.53437 0.05300 0.53548 0.05300 0.54435 0.05400 0.54545 0.05400 0.55543 0.05500 0.55654 0.05500 0.56319 0. "Image_1.txt" 1L, 17458B

[shersoni610]

I was hoping GroundedSAM to give a bounding box.

[capjamesg]

Grounded SAM is a hybrid model that runs Grounding DINO first then SAM, a segmentation model. The output is thus in segmentation form. You can load the values after the first 0 (the class ID) into supervision to get bounding boxes. See https://blog.roboflow.com/convert-bboxes-masks-polygons/#how-to-convert-a-mask-to-bounding-box-mask-to-xyxy for more information.

[capjamesg]

If you want exclusively bounding boxes, autodistill-grounding-dino might be more appropriate (it is also faster).

[csv610]

Hello  I was wondering if yolo8 training model will accept mask as well as bounding box. So far I assuming that all the models listed in autodistill work with boxes.

Sent from Yahoo Mail for iPhone

On Friday, November 3, 2023, 10:41 AM, James @.***> wrote:

Grounded SAM is a hybrid model that runs Grounding DINO first then SAM, a segmentation model. The output is thus in segmentation form. You can load the values after the first 0 (the class ID) into supervision to get bounding boxes. See https://blog.roboflow.com/convert-bboxes-masks-polygons/#how-to-convert-a-mask-to-bounding-box-mask-to-xyxy for more information.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

[capjamesg]

You can train a YOLOv8 segmentation model with Autodistill. See https://github.com/autodistill/autodistill-yolov8?tab=readme-ov-file#choosing-a-task for guidance on how to specify that you want to train a segmentation model. You will need to choose a base model that supported segmentation, like Grounded SAM or FastSAM. See https://github.com/autodistill/autodistill?tab=readme-ov-file#-available-models for more information about supported models.

[csv610]

Perhaps, I asked tge wrong question. I was asking for object detection model.

Sent from Yahoo Mail for iPhone

On Friday, November 3, 2023, 3:43 PM, James @.***> wrote:

You can train a YOLOv8 segmentation model with Autodistill. See https://github.com/autodistill/autodistill-yolov8?tab=readme-ov-file#choosing-a-task for guidance on how to specify that you want to train a segmentation model. You will need to choose a base model that supported segmentation, like Grounded SAM or FastSAM. See https://github.com/autodistill/autodistill?tab=readme-ov-file#-available-models for more information about supported models.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

[capjamesg]

In that case, I recommend Autodistill Grounding DINO. Grounding DINO is a zero-shot object detection model that you can use to label images with bounding boxes.

[csv610]

Hello. I am wondering if grounding Dino is great then what could be reason to include other models such as owlv2, detic, samclip etc.

Sent from Yahoo Mail for iPhone

On Saturday, November 4, 2023, 11:46 PM, James @.***> wrote:

In that case, I recommend Autodistill Grounding DINO. Grounding DINO is a zero-shot object detection model that you can use to label images with bounding boxes.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

[capjamesg]

Good question! Every model has its strengths and weaknesses. We prioritize implementing state-of-the-art models. We expect the library of models to grow as the state of the art gets better.

We encourage people to try different models to see which one works well for their use case. For example, OWLv2 and Grounding DINO are both impressive models capable of identifying a range of objects, but there isn't an objective "best". We often. recommend Grounding DINO as a starting point for object detection, but we haven't noted this in our documentation. We are discussing how best to accomplish this.

[dilwolf]

If you want exclusively bounding boxes, autodistill-grounding-dino might be more appropriate (it is also faster).

Thanks a lot for saving my time! Since it is my first time using autodistill, I had the issue that converting polygon to bbox. Because the getting bbox was just enough for my case. I used autodistill-grounding-dino. This suggestion really helped.

[dilwolf]

If you want exclusively bounding boxes, autodistill-grounding-dino might be more appropriate (it is also faster).

By the way, the accuracy of autodistill-grounding-dino is the same as the accuracy of autodistill-grounding-sam for object detection, right?