Thread by @paulr_rohan

Some tips to increase the Performance of Your Object Detection Models - A trhead
1/n
#computervision #pytorch #deeplearning #deeplearningai #100daysofmlcode #neuralnetworks #machinelearning #datascience #pythonprogramming #python #AI #ArtificialIntelligence #pythonprogramming

2/n
Following tips may boost model performance across different network structures with up to 5% (mAP or mean Average Precision) without increasing computational costs in any way.

#computervision #pytorch #deeplearning #deeplearningai #100daysofmlcode #neuralnetworks #AI

3/n
Visually Coherent Image Mix-up for Object Detection. This has already been proven to be successful in lessening adversarial fears in network classification after testing it on COCO 2017 and PASCAL datasets with YOLOv3 models.
#computervision #pytorch

4/n
Classification Head Label Smoothening

Existing models apply Softmax technique to compute a probability distribution for classes. But there’s a risk of the model becoming too confident in its predictions which can result to over-fitting.
#computervision

5/n
One possible solution to this is to relax our confidence on the labels. For instance, we can slightly lower the loss target values from 1 to, say, 0.9.
#computervision #pytorch #deeplearning #deeplearningai #100daysofmlcode #neuralnetworks #AI #ArtificialIntelligence

6/n
And naturally, we increase the target value of 0 for the others slightly as such. This idea is called label smoothing.

#computervision #pytorch #deeplearning #deeplearningai #100daysofmlcode #neuralnetworks #machinelearning #datascience #pythonprogramming #python #AI

7/n
Data Pre-processing (Mixed Results) like,
• A. Random geometry transformation for random cropping (with constraints), random expansion, random horizontal flip and random resize (with random interpolation).
#computervision #pytorch #deeplearning #deeplearningai

8/n
• B. Random color jittering for brightness, hue, saturation, and contrast
#computervision #pytorch #deeplearning #deeplearningai #100daysofmlcode #neuralnetworks #machinelearning #datascience #pythonprogramming #python #AI #ArtificialIntelligence #pythonprogramming

9/n
Training Scheduler Revamping (+1.44% mAP Boost)

#computervision #pytorch #deeplearning #deeplearningai #100daysofmlcode #neuralnetworks #machinelearning #datascience #pythonprogramming #python #AI #ArtificialIntelligence #pythonprogramming

10/n

In model training, step scheduler is the most widely used learning rate schedule. It involves multiplying the learning rate by a constant number below 1 after a number of model iterations.

#computervision #pytorch #deeplearning #deeplearningai #100daysofmlcode

11/n

Synchronized Batch Normalization (SBN)

What is SBN ?

Standard implementations of BN in public frameworks (such as Caffe, MXNet, Torch, TF, PyTorch) are unsynchronized, which means that the data are normalized within each GPU.

#computervision #pytorch #deeplearning

12/n

Therefore the working batch-size of the BN layer is BatchSize/nGPU (batch-size in each GPU).
Hence the solution is, Synchronized Multi GPU BatchNormalization.

#computervision #pytorch #deeplearning #deeplearningai #100daysofmlcode #neuralnetworks

13/n
Random Shapes Training for Single-stage ODN

#computervision #pytorch #deeplearning #deeplearningai #100daysofmlcode #neuralnetworks #machinelearning #datascience #pythonprogramming #python #AI #ArtificialIntelligence #pythonprogramming