I noticed some stones on the (lab-created) price tracker are denoted as “Ideal Proportions” but when I look at the lab report they fall outside both GIA EX/AGS 0.
An example is James Allen SKU:7203029. The IGI report says table 58%, crown 34.9, pavillion 40.2. When I plug that into the cut estimator it falls in GIA VG/AGS VG.
In fact, not a single IGI report I’ve requested for lab grown diamonds fall in either GIA EX/AGS 0. Am I doing something wrong?!
Thank you!
Hi Justin,
Thank you for your observations. It identifies the limits of machine learning.
For James Allen, I evaluate cut quality and optical symmetry using computer vision to look for 8 crisp arrows (https://www.diamondscreener.com/education/evaluating-round-diamond-cut-quality-using-computer-vision/). My vision-based classifier was trained on natural EX diamonds, the vast majority of which were graded by GIA. GIA EX does not include the proportions of the example diamond you gave, so why I apply my classifier to GIA EX diamonds, such a diamond would never appear.
For lab diamonds, the vast majority are graded by IGI and IGI’s EX grade can include a diamond with a pavilion angle of 40.2. However, my computer vision algorithm still sees that it has 8 strong arrows, which you can confirm in the images. I think 7203029 has nice optical symmetry, so then it’s a question of whether to go by the paper or by what you see.
Interestingly, in the IGI subset of my natural diamond training set, only about 30% of the IGI diamonds that my vision classifier said had ideal proporitons had a pavilion angle < 40.6. I would have guessed that percentage would carry over to the lab diamonds (unless natural and lab IGI diamonds don’t follow the same prior distribution). How many diamonds have you requested the IGI lab reports for? I’m a bit surprised that zero of them qualified.
I just pushed out a new update to the code that increases the score cutoff threshold to try to boost the sensitivity. Maybe give that a try?
I enjoyed digging into this issue.
Thanks,
Joe
Here is a link for the one I’m talking about: https://www.jamesallen.com/loose-diamonds/round-cut/1.26-carat-f-color-vvs2-clarity-ideal-cut-sku-7325705
Upon second glance the last proportions I listed are actually AGS 0, and if you round up the pavillion angle to 40.6 it is also GIA EX. So perhaps I made a poor decision, but I didn’t find that one as visually symmetrical or pleasing: https://www.jamesallen.com/loose-diamonds/round-cut/1.25-carat-e-color-vs1-clarity-ideal-cut-sku-7475985
I’m happy with the one I picked, I just hope it doesn’t disappoint in person!
Also, perhaps of interest to you, the shallowest pavillion of the ones I listed is the only other one your computer vision is identifying as “Ideal Proportions”. 56.5% table 35.5 crown 40.0 pavillion (61.3% total depth) is SKU 7503054.
I like your choice (7325705) more. The optical symmetry is cleaner because the arrows form an 8 point star. 7475985 doesn’t have that and the table isn’t as crisp, as you’ve noticed.
Your choice is nice, don’t worry about it. Cut quality doesn’t drop off a cliff at the boundaries. You’re right on the edge of AGS Ideal if you round to even numbers like GIA would have on their reports.
I think a vision based approach can identify diamonds that have similar visual appearance based on the cutter’s line (Figure 23, https://cdn.ymaws.com/americangemsociety.site-ym.com/resource/resmgr/docs/AGSLab/4661436797413.pdf). So even though it might not match AGS’ categorization, it should have similar visual appearance and light performance when you consider the continuous distribution rather than their cutoffs.
Thank you for the reply! Here are the reports I have (table crown pavillion):
57% 34.1 40.5
56.5% 33.6 40.3
57.5% 33.4 40.4
58% 34.9 40.2
56.5% 35.5 40
55% 35.7 40.1
55.5% 35.8 40.5
I ultimately purchased the first one, which is sku 7325705 and was also identified as “ideal proportions” by your algorithm. My untrained eyes agree with your computer vision based on the image/video. I realize it is on the shallow side so I hope there aren’t obstruction issues in person/in real life. My understanding is the 40.5 pavillion might get rounded up to 40.6 if it was GIA graded so I will just tell myself it is GIA EX and just 0.1 degrees off being also AGS Ideal 😉 I hope the visual symmetry I see in the photos makes up for any of these downfalls. Ultimately I was in a crunch and pretty committed to getting a stone in the 7mm diameter range, so I hope this was a good purchase for $3,800.