Hi all,
\nPlease share your best score with a single model so far.
", "messages": [ { "id": "2247337", "postDate": "05/05/2023 22:36:18", "content": "Hi all,
\nPlease share your best score with a single model so far.
", "rawMarkdown": "Hi all,\n\nPlease share your best score with a single model so far.", "votes": null }, { "id": "2247339", "postDate": "05/05/2023 22:38:44", "content": "Mine so far 0.80
", "rawMarkdown": "Mine so far 0.80", "votes": null }, { "id": "2247864", "postDate": "05/06/2023 10:14:15", "content": "0.79 by now, efficientnet b1 and b3 (either has .79 as single model submission) both were pretrained on 21 / 22 data and then trained using cut-mix augmentation.
\nMaybe you can elaborate a bit on your model @tjamali ?
\nWould really appreciate it!
Best,
\nJan
single fold single model 0.81
", "rawMarkdown": "single fold single model 0.81", "votes": null }, { "id": "2248549", "postDate": "05/07/2023 01:41:17", "content": "I used the SED model by @hidehisaarai1213. Last year, I also used this model. You can also add more to it by inspiring from the solution of the top winners of the previous Bird Competitions.
\nYou can be among the bronze medals easily (even silver medals) if you use an ensemble of your best single models. If you haven't seen this post, take a look at it. Thanks to @leonshangguan, you can speed things up.
I hope that I can be among the gold medals. Of course, it's a dream now π but I am trying.
", "rawMarkdown": "I hope that I can be among the gold medals. Of course, it's a dream now π but I am trying.", "votes": null }, { "id": "2248970", "postDate": "05/07/2023 11:28:21", "content": "mine so far 0.77, trained on 23 dataset only
", "rawMarkdown": "mine so far 0.77, trained on 23 dataset only", "votes": null }, { "id": "2250982", "postDate": "05/09/2023 00:55:44", "content": "0.80 very close to 0.81
", "rawMarkdown": "0.80 very close to 0.81", "votes": null }, { "id": "2265446", "postDate": "05/19/2023 08:50:05", "content": "Simple CNN (eca_nfnet_l0) 0.80 close to 0.81 [CV 90-10 Split 0.86 AP and 0.94 CMAP] (Inference time 100 mins)
\nSimple CNN (regnety_008) 0.80 [CV 90-10 Split 0.858 AP and 0.94 CMAP] (Inference time 30 mins)
Didn't do ensemble till now. Will do before competition ends.
", "rawMarkdown": "Simple CNN (eca_nfnet_l0) 0.80 close to 0.81 [CV 90-10 Split 0.86 AP and 0.94 CMAP] (Inference time 100 mins)\nSimple CNN (regnety_008) 0.80 [CV 90-10 Split 0.858 AP and 0.94 CMAP] (Inference time 30 mins)\n\nDidn't do ensemble till now. Will do before competition ends.", "votes": null }, { "id": "2267435", "postDate": "05/20/2023 22:41:27", "content": "0.82 close to 0.81
", "rawMarkdown": "0.82 close to 0.81", "votes": null }, { "id": "2269867", "postDate": "05/22/2023 18:57:24", "content": "In @leonshangguan post, does the speed up come only from the ThreadPoolExecuter?
", "rawMarkdown": "In @leonshangguan post, does the speed up come only from the ThreadPoolExecuter?", "votes": null }, { "id": "2270057", "postDate": "05/23/2023 00:37:45", "content": "Plus num_workers in dataloader
", "rawMarkdown": "Plus num_workers in dataloader", "votes": null }, { "id": "2273893", "postDate": "05/25/2023 13:41:22", "content": "Mine so far 0.80
", "votes": null, "replies": [] }, { "id": 2247864, "author_name": "janbrederecke", "author_url": "", "post_date": "05/06/2023 10:14:15", "content": "0.79 by now, efficientnet b1 and b3 (either has .79 as single model submission) both were pretrained on 21 / 22 data and then trained using cut-mix augmentation.
\nMaybe you can elaborate a bit on your model @tjamali ?
\nWould really appreciate it!
Best,
\nJan
I used the SED model by @hidehisaarai1213. Last year, I also used this model. You can also add more to it by inspiring from the solution of the top winners of the previous Bird Competitions.
\nYou can be among the bronze medals easily (even silver medals) if you use an ensemble of your best single models. If you haven't seen this post, take a look at it. Thanks to @leonshangguan, you can speed things up.
In @leonshangguan post, does the speed up come only from the ThreadPoolExecuter?
", "votes": null, "replies": [ { "id": 2270057, "author_name": "tjamali", "author_url": "", "post_date": "05/23/2023 00:37:45", "content": "Plus num_workers in dataloader
", "votes": null, "replies": [] } ] } ] } ] }, { "id": 2248028, "author_name": "leonshangguan", "author_url": "", "post_date": "05/06/2023 13:07:03", "content": "single fold single model 0.81
", "votes": null, "replies": [ { "id": 2248552, "author_name": "tjamali", "author_url": "", "post_date": "05/07/2023 01:42:44", "content": "I hope that I can be among the gold medals. Of course, it's a dream now π but I am trying.
", "votes": null, "replies": [] } ] }, { "id": 2248970, "author_name": "doubleastrong", "author_url": "", "post_date": "05/07/2023 11:28:21", "content": "mine so far 0.77, trained on 23 dataset only
", "votes": null, "replies": [] }, { "id": 2250982, "author_name": "infiniteemo", "author_url": "", "post_date": "05/09/2023 00:55:44", "content": "0.80 very close to 0.81
", "votes": null, "replies": [] }, { "id": 2265446, "author_name": "salmanahmedtamu", "author_url": "", "post_date": "05/19/2023 08:50:05", "content": "Simple CNN (eca_nfnet_l0) 0.80 close to 0.81 [CV 90-10 Split 0.86 AP and 0.94 CMAP] (Inference time 100 mins)
\nSimple CNN (regnety_008) 0.80 [CV 90-10 Split 0.858 AP and 0.94 CMAP] (Inference time 30 mins)
Didn't do ensemble till now. Will do before competition ends.
", "votes": null, "replies": [] }, { "id": 2267435, "author_name": "shigemitsutomizawa", "author_url": "", "post_date": "05/20/2023 22:41:27", "content": "0.82 close to 0.81
", "votes": null, "replies": [ { "id": 2273893, "author_name": "shigemitsutomizawa", "author_url": "", "post_date": "05/25/2023 13:41:22", "content": "