Did you already try our image based object recognition map creation? We launched this with our SDK 8 beta and it will replace the video based approach. This new and improved way to generate object maps will also increase the quality in terms of e.g. 360° maps.
So I strongly advice to check out the SDK 8 beta and the image based creation process.
Our team has created several object recognition catalogs via video. We have pretty consistently seen that the object will be better recognized on the initial side filmed in the video. Do you have any tips regarding how one make recognition work equally as well on both sides of an object?