Jump to content
Maestronet Forums

AlexChandler

New Members
  • Posts

    8
  • Joined

  • Last visited

Recent Profile Visitors

The recent visitors block is disabled and is not being shown to other users.

AlexChandler's Achievements

New Member

New Member (1/5)

  1. There is a chance that there are multiple errors in my code. I’ve been working on this project for under two months, and have had zero supervision. Again my work is both untested and unfinished. Once I post the instrument database, however, I feel a good amount of justified skepticism will drop just because people can check for themselves what information I used. As to the question of whether my validation data stole was stolen from the training data, I believe the answer is no. I used Keras preprocessing data train and test splitting. But then again, I may be wrong. My CNN model will likely chance to VG16, xCeption, or a modification of one of the two. If you would like to check for yourself, I posed at least half of my code on github (including the machine learning code). If I did make any glaring mistakes, please let me know and I will fix them. https://github.com/achandlr/Violin-Project-Code
  2. Yes a logarithmic scale would be much more helpful. With instrument valued skewed as much as they are, a change in the scale is fully warranted.
  3. Perhaps I was too harsh as well. I am planning on building a country of origin identifier for Italy, Germany, and France, and England, or other. Unfortunately the other category is unavoidable because one needs at least 1000 pieces of data to build a high accuracy model. Most likely the machine learning side of the project will be the least useful part because it by itself will not be more accurate than 80 percent (although fo be fair I see instruments misidentified either intentionally or unintentionally all the time). I just thought the machine learning side would be a fun thing to try and a good challenge to learn from. If I have time, my hope is to do a mix of photo analysis along with maker name and other factors. For example, take the maker Antonio Stradivari. One could plot a distribution of his instruments value over the the time that they were built, combining the predicted value based off instrument type, year made, and maker name, to the estimated quality of violin compared to other instruments of the same maker. Lets take Testore for example. I doubt my machine learning code values his instruments as high as they should be, because honestly, a lot of them don’t look as nice as they sound. Ok. So one way I could fix this is use my algorithm to predict the price of all Testore instruments and then scale each instrument to their actual values. Not sure if it will work, but it could solve the model vs real instrument price dilemma when a 30,000 dollar copy looks identical to a 500,000 dollar instrument. This is partially cheating if my sole purpose is to only use photos, but I think it would have the potential to work really well for makers with over 10 instruments. I’m not sure how well this method would scale down to makers with fewer instruments in my database. My real hope others can at least find use of my database.
  4. Maybe. I think I will make another post once I am ready to post everything in an accessible format. Who knows who will use it but I hope someone will find it handy for whatever they need. I think I’ll keep the photo data out of it or at a reduces number of pixels just because than the file would be over 50 gigabytes.
  5. The quality of how an instrument looks is a predictive measure for the value of an instrument. It is one of many measures, like maker name, but I suggest you learn more about Convolutional neural networks with multiple inputs before comparing them to 19th century phrenology.
  6. I hear you and share your skepticism in such a model working. In my opinion, artificial intelligence will never do at good a job as predicting the price Of instrument than a trained lutier, but for it to work for remotely is pretty cool. The algorithm is a kind of a deep convolution of neural network, but I am not fully satisfied with it so I am trying different models. For anyone with tech knowledge, the algorithms are VG16 and xCeption. Similar architecture’s are used to categorize faces, although the power of similar architectures far exceeds facial recognition. Currently, my model works well at distinguishing a shitty instrument from a good instrument, but it does not do a good job of distinguishing between a good instrument and a great instrument. I think the more useful part of my project is inflation data for different types of instruments. I am an amateur cellist, and have long been interested in how instruments appreciate over time.
  7. Obviously some antique instruments are to have repairs. What troubles me is the back damage and the extensive amount of scratches. And I understand fine english cellos from the early 1800s can go for up to 200K. I saw a Kennedy Cello go for 180K this year. My question is to how it will sound with all the damage.
  8. Brompton’s auction started and I was looking at this cello. To me, it appears to have been once a very nice instrument, but now I fear it’s condition, particularly the soundpost patch and all the scratches on its back, will make the cello no longer sound very nice. What do y’all think? Is it worth the estimate and would it be a good investment? Will it sound like a Banks cello in good condition, or is there enough damage to severely hurt the sound? https://www.bromptons.co/auction/22nd-june-2020/lots/157-a-good-english-cello-by-james-and-henry-banks-salisbury-1800.html
×
×
  • Create New...