AIM: TO IDENTIFY A APPROPRIATE METHOD TO GENERATE SYNTHETIC DATA WHICH CAN BE USED TO TRAIN PRETRAINED CLASSIFICATION MODELS.

Currently the synthetic data is generated using Unity3D, we plan on moving it to a more light weight 3D engine.

The 3d models are obtained from ShapeNet Dataset

For prototyping selected classes of random objects are spawned far away from the Camera, while the target(knife in the below example) object is spawned infront. For further simplicity we are ignoring complex real world scenarios also, which we plan on implementing further down the road.

Here are two train images for a knife the system generated:

• Benchmarking the synthetic images.

• Improve the Object spawning technique to get all the areas.

• Change the Background of the walls, randomized textures.

• The intraction of lighting to different objects should be made more realistic.

• Create a GAN which reduces test loss by appropriately synthesysing the background pixels to the different poses.