ssd_keras
FastMOT
Our great sponsors
ssd_keras | FastMOT | |
---|---|---|
4 | 2 | |
1,846 | 1,095 | |
- | - | |
0.0 | 0.0 | |
about 2 years ago | over 2 years ago | |
Python | Python | |
Apache License 2.0 | MIT License |
Stars - the number of stars that a project has on GitHub. Growth - month over month growth in stars.
Activity is a relative number indicating how actively a project is being developed. Recent commits have higher weight than older ones.
For example, an activity of 9.0 indicates that a project is amongst the top 10% of the most actively developed projects that we are tracking.
ssd_keras
-
Failed to get convolution algorithm. This is probably because cuDNN failed to initialize,
In Tensorflow/ Keras when running the code from https://github.com/pierluigiferrari/ssd_keras, use the estimator: ssd300_evaluation. I received this error.
-
Shared weights between different implementations
Yeah, the order of axes was different between those 2. Another guy used https://github.com/pierluigiferrari/ssd_keras https://github.com/uhfband/keras2caffe/blob/master/keras2caffe/convert.py probably not much actual use but maybe some more reassurance?
-
Simplest way to deploy Keras NN model into C++?
Don't know about simplest, but we either used caffe or tensorrt, it is maybe a bit difficult to use but I'd actually say simple fast GPU inference is what it's geared towards. There is a keras -> caffe converter https://github.com/pierluigiferrari/ssd_keras here, I think. Caffe is a c++ lib, typical, with dependencies and all. I've never heard anything of tensorflow running on c++. But with tensorrt you should get an "artifact" that you'd load, no matter where it comes from
-
ValueError: Layer model expects 1 input(s), but it received 2 input tensors. Help?
Tensorflow V1 Keras code (original repo): Github Repo
FastMOT
- Does Multi Object Tracking work better (precision/recall) on videos than jury rigging a SOTA image object detection to work on videos?
-
Assign ID and track moving object with optical flow
On failure, you can try using a re-identification methods like FastReid: https://github.com/JDAI-CV/fast-reid in combination with your detector. A good pipeline that combines everything you seem to need is here: https://github.com/GeekAlexis/FastMOT. It uses a combination of Yolov4 (detector) + Kalman filters, Optical flow (tracker) and FastReid (re-identification)
What are some alternatives?
layout-parser - A Unified Toolkit for Deep Learning Based Document Image Analysis
multi-object-tracker - Multi-object trackers in Python
cppflow - Run TensorFlow models in C++ without installation and without Bazel
Deep-SORT-YOLOv4 - People detection and optional tracking with Tensorflow backend.
zero-shot-object-tracking - Object tracking implemented with the Roboflow Inference API, DeepSort, and OpenAI CLIP.
ByteTrack - [ECCV 2022] ByteTrack: Multi-Object Tracking by Associating Every Detection Box
efficientnet-lite-keras - Keras reimplementation of EfficientNet Lite.
fast-reid - SOTA Re-identification Methods and Toolbox
Mask-RCNN-TF2 - Mask R-CNN for object detection and instance segmentation on Keras and TensorFlow 2.0
TFJS-object-detection - Real-time custom object detection in the browser using tensorflow.js
a-PyTorch-Tutorial-to-Object-Detection - SSD: Single Shot MultiBox Detector | a PyTorch Tutorial to Object Detection