Git Product home page Git Product logo

Comments (5)

mikel-brostrom avatar mikel-brostrom commented on June 20, 2024 2

Segmentation model vs. object detection model on MOT metrics

You will need to make custom modifications. Pass both bboxes and masks to the tracker and mask the crops generated with bboxes using the masks:

for box, mask in zip(bbox_xywhs, masks):
x1, y1, x2, y2 = self._xywh_to_xyxy(box)
m = mask[y1:y2, x1:x2]
crop = ori_img[y1:y2, x1:x2]
# grey out everything that is not the detected object's mask
masked_crop = np.where(m[...,None] == 0, (114, 114, 114), crop).astype(np.uint8)

Segmentation model vs. object detection model on MOT metrics

Segmentation models generate both segmentations and bboxes. Only pass the bboxes πŸ˜„

from yolo_tracking.

mikel-brostrom avatar mikel-brostrom commented on June 20, 2024 2

could you briefly explain the specific implementation of "Masked detection crops vs. regular detection crops for ReID feature extraction,"

Everything you need to know regarding this is in this branch:

for box, mask in zip(bbox_xywhs, masks):
x1, y1, x2, y2 = self._xywh_to_xyxy(box)
m = mask[y1:y2, x1:x2]
crop = ori_img[y1:y2, x1:x2]
# grey out everything that is not the detected object's mask
masked_crop = np.where(m[...,None] == 0, (114, 114, 114), crop).astype(np.uint8)

how it differs in code and approach from the "Segmentation model vs. object detection model on MOT metrics"?

Segmentation models also generate bboxes. These bboxes can be passed instead of the object detection bboxes.

from yolo_tracking.

carry-all-coder avatar carry-all-coder commented on June 20, 2024 1

Got it! So the "Segmentation model vs. object detection model on MOT metrics" experiment refers to model performance between yolov5m and yolov5m-seg(use box output). And the "Masked detection crops vs. regular detection crops for ReID feature extraction" experiment refers to performance comparison between yolov5m-seg and yolov5m-seg with masked refined features, which is related to the code you mentioned.
Thank you for your enlightening response!

from yolo_tracking.

carry-all-coder avatar carry-all-coder commented on June 20, 2024

Dear Mike,

I am deeply grateful for your quick and detailed response to my earlier question. It provided valuable insights!

Another question, could you briefly explain the specific implementation of "Masked detection crops vs. regular detection crops for ReID feature extraction," particularly focusing on how it differs in code and approach from the "Segmentation model vs. object detection model on MOT metrics"? I'm also curious about why segmentation enhances tracking in ReID(Masked detection crops vs. regular detection crops for ReID feature extraction) but not as much in the MOT metrics experiment(Segmentation model vs. object detection model on MOT metrics).

Thank you once again for your time and expertise!

from yolo_tracking.

mikel-brostrom avatar mikel-brostrom commented on June 20, 2024

Exactly. Good luck with your experiments πŸ˜„

from yolo_tracking.

Related Issues (20)

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    πŸ–– Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. πŸ“ŠπŸ“ˆπŸŽ‰

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google ❀️ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.